[ { "id": "advbox", "directoryUrl": "https://awesomemlsecops.com/tools/entries/advbox/", "categoryUrl": "https://awesomemlsecops.com/tools/adversarial-ml/", "category": "adversarial-ml", "name": "AdvBox", "sourceUrl": "https://github.com/advboxes/AdvBox", "description": "Advbox is a toolbox to generate adversarial examples that fool neural networks in PaddlePaddle、PyTorch、Caffe2、MxNet、Keras、TensorFlow", "directoryDescription": "AdvBox is included in the Awesome MLSecOps Adversarial Machine Learning directory. The community-maintained README describes it as: “Advbox is a toolbox to generate adversarial examples that fool neural networks in PaddlePaddle、PyTorch、Caffe2、MxNet、Keras、TensorFlow.” Its MLSecOps relevance is the controlled evaluation of evasion, poisoning, extraction, inversion, privacy attacks, or robustness under an explicit attacker model. The linked first-party source is the advboxes/AdvBox repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat-model fit, Modality and framework support, Adaptive attack evaluation, and Experiment reproducibility. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "adversarial-machine-learning-ctf", "directoryUrl": "https://awesomemlsecops.com/tools/entries/adversarial-machine-learning-ctf/", "categoryUrl": "https://awesomemlsecops.com/tools/adversarial-ml/", "category": "adversarial-ml", "name": "Adversarial Machine Learning CTF", "sourceUrl": "https://github.com/arturmiller/adversarial_ml_ctf", "description": "CTF challenge demonstrating adversarial-image attacks against neural-network image classifiers", "directoryDescription": "Adversarial Machine Learning CTF is included in the Awesome MLSecOps Adversarial Machine Learning directory. The community-maintained README describes it as: “CTF challenge demonstrating adversarial-image attacks against neural-network image classifiers.” Its MLSecOps relevance is the controlled evaluation of evasion, poisoning, extraction, inversion, privacy attacks, or robustness under an explicit attacker model. The linked first-party source is the arturmiller/adversarial_ml_ctf repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat-model fit, Modality and framework support, Adaptive attack evaluation, and Experiment reproducibility. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "adversarial-robustness-toolbox", "directoryUrl": "https://awesomemlsecops.com/tools/entries/adversarial-robustness-toolbox/", "categoryUrl": "https://awesomemlsecops.com/tools/adversarial-ml/", "category": "adversarial-ml", "name": "Adversarial Robustness Toolbox", "sourceUrl": "https://github.com/IBM/adversarial-robustness-toolbox", "description": "Library of defense methods for ML models against adversarial attacks", "directoryDescription": "Adversarial Robustness Toolbox is included in the Awesome MLSecOps Adversarial Machine Learning directory. The community-maintained README describes it as: “Library of defense methods for ML models against adversarial attacks.” Its MLSecOps relevance is the controlled evaluation of evasion, poisoning, extraction, inversion, privacy attacks, or robustness under an explicit attacker model. The linked first-party source is the IBM/adversarial-robustness-toolbox repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat-model fit, Modality and framework support, Adaptive attack evaluation, and Experiment reproducibility. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "advertorch", "directoryUrl": "https://awesomemlsecops.com/tools/entries/advertorch/", "categoryUrl": "https://awesomemlsecops.com/tools/adversarial-ml/", "category": "adversarial-ml", "name": "Advertorch", "sourceUrl": "https://github.com/BorealisAI/advertorch", "description": "Python toolbox for adversarial robustness research", "directoryDescription": "Advertorch is included in the Awesome MLSecOps Adversarial Machine Learning directory. The community-maintained README describes it as: “Python toolbox for adversarial robustness research.” Its MLSecOps relevance is the controlled evaluation of evasion, poisoning, extraction, inversion, privacy attacks, or robustness under an explicit attacker model. The linked first-party source is the BorealisAI/advertorch repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat-model fit, Modality and framework support, Adaptive attack evaluation, and Experiment reproducibility. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "augly", "directoryUrl": "https://awesomemlsecops.com/tools/entries/augly/", "categoryUrl": "https://awesomemlsecops.com/tools/adversarial-ml/", "category": "adversarial-ml", "name": "AugLy", "sourceUrl": "https://github.com/facebookresearch/AugLy", "description": "Multimodal data-augmentation library for evaluating and improving model robustness; it is not an adversarial-attack generator", "directoryDescription": "AugLy is included in the Awesome MLSecOps Adversarial Machine Learning directory. The community-maintained README describes it as: “Multimodal data-augmentation library for evaluating and improving model robustness; it is not an adversarial-attack generator.” Its MLSecOps relevance is the controlled evaluation of evasion, poisoning, extraction, inversion, privacy attacks, or robustness under an explicit attacker model. The linked first-party source is the facebookresearch/AugLy repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat-model fit, Modality and framework support, Adaptive attack evaluation, and Experiment reproducibility. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "cleverhans", "directoryUrl": "https://awesomemlsecops.com/tools/entries/cleverhans/", "categoryUrl": "https://awesomemlsecops.com/tools/adversarial-ml/", "category": "adversarial-ml", "name": "CleverHans", "sourceUrl": "https://github.com/cleverhans-lab/cleverhans", "description": "A library of adversarial examples and defenses for machine learning models", "directoryDescription": "CleverHans is included in the Awesome MLSecOps Adversarial Machine Learning directory. The community-maintained README describes it as: “A library of adversarial examples and defenses for machine learning models.” Its MLSecOps relevance is the controlled evaluation of evasion, poisoning, extraction, inversion, privacy attacks, or robustness under an explicit attacker model. The linked first-party source is the cleverhans-lab/cleverhans repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat-model fit, Modality and framework support, Adaptive attack evaluation, and Experiment reproducibility. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "copycat-cnn", "directoryUrl": "https://awesomemlsecops.com/tools/entries/copycat-cnn/", "categoryUrl": "https://awesomemlsecops.com/tools/adversarial-ml/", "category": "adversarial-ml", "name": "Copycat CNN", "sourceUrl": "https://github.com/jeiks/Stealing_DL_Models", "description": "Proof of concept for training a surrogate convolutional neural network from black-box queries", "directoryDescription": "Copycat CNN is included in the Awesome MLSecOps Adversarial Machine Learning directory. The community-maintained README describes it as: “Proof of concept for training a surrogate convolutional neural network from black-box queries.” Its MLSecOps relevance is the controlled evaluation of evasion, poisoning, extraction, inversion, privacy attacks, or robustness under an explicit attacker model. The linked first-party source is the jeiks/Stealing_DL_Models repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat-model fit, Modality and framework support, Adaptive attack evaluation, and Experiment reproducibility. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "deep-pwning", "directoryUrl": "https://awesomemlsecops.com/tools/entries/deep-pwning/", "categoryUrl": "https://awesomemlsecops.com/tools/adversarial-ml/", "category": "adversarial-ml", "name": "Deep Pwning", "sourceUrl": "https://github.com/cchio/deep-pwning", "description": "Deep-pwning is a lightweight framework for experimenting with machine learning models with the goal of evaluating their robustness against a motivated adversary", "directoryDescription": "Deep Pwning is included in the Awesome MLSecOps Adversarial Machine Learning directory. The community-maintained README describes it as: “Deep-pwning is a lightweight framework for experimenting with machine learning models with the goal of evaluating their robustness against a motivated adversary.” Its MLSecOps relevance is the controlled evaluation of evasion, poisoning, extraction, inversion, privacy attacks, or robustness under an explicit attacker model. The linked first-party source is the cchio/deep-pwning repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat-model fit, Modality and framework support, Adaptive attack evaluation, and Experiment reproducibility. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "foolbox", "directoryUrl": "https://awesomemlsecops.com/tools/entries/foolbox/", "categoryUrl": "https://awesomemlsecops.com/tools/adversarial-ml/", "category": "adversarial-ml", "name": "Foolbox", "sourceUrl": "https://github.com/bethgelab/foolbox", "description": "Python toolbox for creating and evaluating adversarial attacks and defenses", "directoryDescription": "Foolbox is included in the Awesome MLSecOps Adversarial Machine Learning directory. The community-maintained README describes it as: “Python toolbox for creating and evaluating adversarial attacks and defenses.” Its MLSecOps relevance is the controlled evaluation of evasion, poisoning, extraction, inversion, privacy attacks, or robustness under an explicit attacker model. The linked first-party source is the bethgelab/foolbox repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat-model fit, Modality and framework support, Adaptive attack evaluation, and Experiment reproducibility. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "knockoffnets", "directoryUrl": "https://awesomemlsecops.com/tools/entries/knockoffnets/", "categoryUrl": "https://awesomemlsecops.com/tools/adversarial-ml/", "category": "adversarial-ml", "name": "Knockoffnets", "sourceUrl": "https://github.com/tribhuvanesh/knockoffnets", "description": "Proof of concept for training a functional surrogate from black-box model queries", "directoryDescription": "Knockoffnets is included in the Awesome MLSecOps Adversarial Machine Learning directory. The community-maintained README describes it as: “Proof of concept for training a functional surrogate from black-box model queries.” Its MLSecOps relevance is the controlled evaluation of evasion, poisoning, extraction, inversion, privacy attacks, or robustness under an explicit attacker model. The linked first-party source is the tribhuvanesh/knockoffnets repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat-model fit, Modality and framework support, Adaptive attack evaluation, and Experiment reproducibility. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "mitre-atlas", "directoryUrl": "https://awesomemlsecops.com/tools/entries/mitre-atlas/", "categoryUrl": "https://awesomemlsecops.com/tools/adversarial-ml/", "category": "adversarial-ml", "name": "MITRE ATLAS", "sourceUrl": "https://atlas.mitre.org/", "description": "Knowledge base of adversary tactics and techniques against AI systems; successor to the deprecated Adversarial ML Threat Matrix", "directoryDescription": "MITRE ATLAS is included in the Awesome MLSecOps Adversarial Machine Learning directory. The community-maintained README describes it as: “Knowledge base of adversary tactics and techniques against AI systems; successor to the deprecated Adversarial ML Threat Matrix.” Its MLSecOps relevance is the controlled evaluation of evasion, poisoning, extraction, inversion, privacy attacks, or robustness under an explicit attacker model. The linked first-party source is published at atlas.mitre.org. A technical review should test the project's documented evidence across four criteria: Threat-model fit, Modality and framework support, Adaptive attack evaluation, and Experiment reproducibility. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "mlsploit", "directoryUrl": "https://awesomemlsecops.com/tools/entries/mlsploit/", "categoryUrl": "https://awesomemlsecops.com/tools/adversarial-ml/", "category": "adversarial-ml", "name": "MLSploit", "sourceUrl": "https://github.com/mlsploit/", "description": "Cloud framework for interactive experimentation with adversarial machine learning research", "directoryDescription": "MLSploit is included in the Awesome MLSecOps Adversarial Machine Learning directory. The community-maintained README describes it as: “Cloud framework for interactive experimentation with adversarial machine learning research.” Its MLSecOps relevance is the controlled evaluation of evasion, poisoning, extraction, inversion, privacy attacks, or robustness under an explicit attacker model. The linked first-party source is published at github.com. A technical review should test the project's documented evidence across four criteria: Threat-model fit, Modality and framework support, Adaptive attack evaluation, and Experiment reproducibility. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "model-inversion-attack-toolbox", "directoryUrl": "https://awesomemlsecops.com/tools/entries/model-inversion-attack-toolbox/", "categoryUrl": "https://awesomemlsecops.com/tools/adversarial-ml/", "category": "adversarial-ml", "name": "Model-Inversion-Attack-ToolBox", "sourceUrl": "https://github.com/ffhibnese/Model-Inversion-Attack-ToolBox", "description": "A framework for implementing Model Inversion attacks", "directoryDescription": "Model-Inversion-Attack-ToolBox is included in the Awesome MLSecOps Adversarial Machine Learning directory. The community-maintained README describes it as: “A framework for implementing Model Inversion attacks.” Its MLSecOps relevance is the controlled evaluation of evasion, poisoning, extraction, inversion, privacy attacks, or robustness under an explicit attacker model. The linked first-party source is the ffhibnese/Model-Inversion-Attack-ToolBox repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat-model fit, Modality and framework support, Adaptive attack evaluation, and Experiment reproducibility. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "offensive-ml-playbook", "directoryUrl": "https://awesomemlsecops.com/tools/entries/offensive-ml-playbook/", "categoryUrl": "https://awesomemlsecops.com/tools/adversarial-ml/", "category": "adversarial-ml", "name": "Offensive ML Playbook", "sourceUrl": "https://wiki.offsecml.com/Welcome+to+the+Offensive+ML+Playbook", "description": "Offensive ML Playbook. Notes on machine learning attacks and pentesting", "directoryDescription": "Offensive ML Playbook is included in the Awesome MLSecOps Adversarial Machine Learning directory. The community-maintained README describes it as: “Offensive ML Playbook. Notes on machine learning attacks and pentesting.” Its MLSecOps relevance is the controlled evaluation of evasion, poisoning, extraction, inversion, privacy attacks, or robustness under an explicit attacker model. The linked first-party source is published at wiki.offsecml.com. A technical review should test the project's documented evidence across four criteria: Threat-model fit, Modality and framework support, Adaptive attack evaluation, and Experiment reproducibility. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "openattack", "directoryUrl": "https://awesomemlsecops.com/tools/entries/openattack/", "categoryUrl": "https://awesomemlsecops.com/tools/adversarial-ml/", "category": "adversarial-ml", "name": "OpenAttack", "sourceUrl": "https://github.com/thunlp/OpenAttack", "description": "An Open-Source Package for Textual Adversarial Attack", "directoryDescription": "OpenAttack is included in the Awesome MLSecOps Adversarial Machine Learning directory. The community-maintained README describes it as: “An Open-Source Package for Textual Adversarial Attack.” Its MLSecOps relevance is the controlled evaluation of evasion, poisoning, extraction, inversion, privacy attacks, or robustness under an explicit attacker model. The linked first-party source is the thunlp/OpenAttack repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat-model fit, Modality and framework support, Adaptive attack evaluation, and Experiment reproducibility. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "raze-to-the-ground-query-efficient-adversarial-html-attacks-on-machine-learning-phishing-webpage-detectors", "directoryUrl": "https://awesomemlsecops.com/tools/entries/raze-to-the-ground-query-efficient-adversarial-html-attacks-on-machine-learning-phishing-webpage-detectors/", "categoryUrl": "https://awesomemlsecops.com/tools/adversarial-ml/", "category": "adversarial-ml", "name": "Raze to the Ground: Query-Efficient Adversarial HTML Attacks on Machine-Learning Phishing Webpage Detectors", "sourceUrl": "https://github.com/advmlphish/raze_to_the_ground_aisec23", "description": "Source code of the paper \"Raze to the Ground: Query-Efficient Adversarial HTML Attacks on Machine-Learning Phishing Webpage Detectors\" accepted at AISec '23", "directoryDescription": "Raze to the Ground: Query-Efficient Adversarial HTML Attacks on Machine-Learning Phishing Webpage Detectors is included in the Awesome MLSecOps Adversarial Machine Learning directory. The community-maintained README describes it as: “Source code of the paper \"Raze to the Ground: Query-Efficient Adversarial HTML Attacks on Machine-Learning Phishing Webpage Detectors\" accepted at AISec '23.” Its MLSecOps relevance is the controlled evaluation of evasion, poisoning, extraction, inversion, privacy attacks, or robustness under an explicit attacker model. The linked first-party source is the advmlphish/raze_to_the_ground_aisec23 repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat-model fit, Modality and framework support, Adaptive attack evaluation, and Experiment reproducibility. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "textattack", "directoryUrl": "https://awesomemlsecops.com/tools/entries/textattack/", "categoryUrl": "https://awesomemlsecops.com/tools/adversarial-ml/", "category": "adversarial-ml", "name": "TextAttack", "sourceUrl": "https://github.com/QData/TextAttack", "description": "TextAttack is a Python framework for adversarial attacks, data augmentation, and model training in NLP", "directoryDescription": "TextAttack is included in the Awesome MLSecOps Adversarial Machine Learning directory. The community-maintained README describes it as: “TextAttack is a Python framework for adversarial attacks, data augmentation, and model training in NLP.” Its MLSecOps relevance is the controlled evaluation of evasion, poisoning, extraction, inversion, privacy attacks, or robustness under an explicit attacker model. The linked first-party source is the QData/TextAttack repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat-model fit, Modality and framework support, Adaptive attack evaluation, and Experiment reproducibility. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "textfooler", "directoryUrl": "https://awesomemlsecops.com/tools/entries/textfooler/", "categoryUrl": "https://awesomemlsecops.com/tools/adversarial-ml/", "category": "adversarial-ml", "name": "TextFooler", "sourceUrl": "https://github.com/jind11/TextFooler", "description": "A Model for Natural Language Attack on Text Classification and Inference", "directoryDescription": "TextFooler is included in the Awesome MLSecOps Adversarial Machine Learning directory. The community-maintained README describes it as: “A Model for Natural Language Attack on Text Classification and Inference.” Its MLSecOps relevance is the controlled evaluation of evasion, poisoning, extraction, inversion, privacy attacks, or robustness under an explicit attacker model. The linked first-party source is the jind11/TextFooler repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat-model fit, Modality and framework support, Adaptive attack evaluation, and Experiment reproducibility. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "vger", "directoryUrl": "https://awesomemlsecops.com/tools/entries/vger/", "categoryUrl": "https://awesomemlsecops.com/tools/adversarial-ml/", "category": "adversarial-ml", "name": "VGER", "sourceUrl": "https://github.com/JosephTLucas/vger", "description": "Jupyter Attack framework", "directoryDescription": "VGER is included in the Awesome MLSecOps Adversarial Machine Learning directory. The community-maintained README describes it as: “Jupyter Attack framework.” Its MLSecOps relevance is the controlled evaluation of evasion, poisoning, extraction, inversion, privacy attacks, or robustness under an explicit attacker model. The linked first-party source is the JosephTLucas/vger repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat-model fit, Modality and framework support, Adaptive attack evaluation, and Experiment reproducibility. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "agent-memory-guard", "directoryUrl": "https://awesomemlsecops.com/tools/entries/agent-memory-guard/", "categoryUrl": "https://awesomemlsecops.com/tools/agent-security/", "category": "agent-security", "name": "Agent Memory Guard", "sourceUrl": "https://github.com/OWASP/www-project-agent-memory-guard", "description": "OWASP project providing a runtime defense layer that screens AI agent memory reads and writes against prompt injection, secret leakage, and memory poisoning", "directoryDescription": "Agent Memory Guard is included in the Awesome MLSecOps AI Agent and MCP Security directory. The community-maintained README describes it as: “OWASP project providing a runtime defense layer that screens AI agent memory reads and writes against prompt injection, secret leakage, and memory poisoning.” Its MLSecOps relevance is the security of agent identities, tool permissions, memory, credentials, approval paths, sandboxes, runtime actions, or MCP integrations. The linked first-party source is the OWASP/www-project-agent-memory-guard repository on GitHub. A technical review should test the project's documented evidence across four criteria: Tool-level authorization, Memory and context controls, Sandbox boundaries, and Audit and approval workflows. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "agent-wiz", "directoryUrl": "https://awesomemlsecops.com/tools/entries/agent-wiz/", "categoryUrl": "https://awesomemlsecops.com/tools/agent-security/", "category": "agent-security", "name": "Agent-Wiz", "sourceUrl": "https://github.com/Repello-AI/Agent-Wiz", "description": "Python CLI by Repello AI for extracting agentic workflows from LangChain/LangGraph/CrewAI/AutoGen and running automated threat modeling", "directoryDescription": "Agent-Wiz is included in the Awesome MLSecOps AI Agent and MCP Security directory. The community-maintained README describes it as: “Python CLI by Repello AI for extracting agentic workflows from LangChain/LangGraph/CrewAI/AutoGen and running automated threat modeling.” Its MLSecOps relevance is the security of agent identities, tool permissions, memory, credentials, approval paths, sandboxes, runtime actions, or MCP integrations. The linked first-party source is the Repello-AI/Agent-Wiz repository on GitHub. A technical review should test the project's documented evidence across four criteria: Tool-level authorization, Memory and context controls, Sandbox boundaries, and Audit and approval workflows. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "agentic-security", "directoryUrl": "https://awesomemlsecops.com/tools/entries/agentic-security/", "categoryUrl": "https://awesomemlsecops.com/tools/agent-security/", "category": "agent-security", "name": "Agentic Security", "sourceUrl": "https://github.com/msoedov/agentic_security", "description": "Agentic LLM Vulnerability Scanner / AI red teaming kit", "directoryDescription": "Agentic Security is included in the Awesome MLSecOps AI Agent and MCP Security directory. The community-maintained README describes it as: “Agentic LLM Vulnerability Scanner / AI red teaming kit.” Its MLSecOps relevance is the security of agent identities, tool permissions, memory, credentials, approval paths, sandboxes, runtime actions, or MCP integrations. The linked first-party source is the msoedov/agentic_security repository on GitHub. A technical review should test the project's documented evidence across four criteria: Tool-level authorization, Memory and context controls, Sandbox boundaries, and Audit and approval workflows. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "ironclaw", "directoryUrl": "https://awesomemlsecops.com/tools/entries/ironclaw/", "categoryUrl": "https://awesomemlsecops.com/tools/agent-security/", "category": "agent-security", "name": "IronClaw", "sourceUrl": "https://github.com/IronSecCo/ironclaw", "description": "Self-hosted runtime that sandboxes autonomous AI agents with isolated execution, restricted networking, and human-approval gating", "directoryDescription": "IronClaw is included in the Awesome MLSecOps AI Agent and MCP Security directory. The community-maintained README describes it as: “Self-hosted runtime that sandboxes autonomous AI agents with isolated execution, restricted networking, and human-approval gating.” Its MLSecOps relevance is the security of agent identities, tool permissions, memory, credentials, approval paths, sandboxes, runtime actions, or MCP integrations. The linked first-party source is the IronSecCo/ironclaw repository on GitHub. A technical review should test the project's documented evidence across four criteria: Tool-level authorization, Memory and context controls, Sandbox boundaries, and Audit and approval workflows. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "skill-audit-mcp", "directoryUrl": "https://awesomemlsecops.com/tools/entries/skill-audit-mcp/", "categoryUrl": "https://awesomemlsecops.com/tools/agent-security/", "category": "agent-security", "name": "skill-audit-mcp", "sourceUrl": "https://github.com/eltociear/skill-audit-mcp", "description": "Static security scanner for MCP servers, AI agent skills, and plugins, with SARIF output and CI integration", "directoryDescription": "skill-audit-mcp is included in the Awesome MLSecOps AI Agent and MCP Security directory. The community-maintained README describes it as: “Static security scanner for MCP servers, AI agent skills, and plugins, with SARIF output and CI integration.” Its MLSecOps relevance is the security of agent identities, tool permissions, memory, credentials, approval paths, sandboxes, runtime actions, or MCP integrations. The linked first-party source is the eltociear/skill-audit-mcp repository on GitHub. A technical review should test the project's documented evidence across four criteria: Tool-level authorization, Memory and context controls, Sandbox boundaries, and Audit and approval workflows. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "ai-goat", "directoryUrl": "https://awesomemlsecops.com/tools/entries/ai-goat/", "categoryUrl": "https://awesomemlsecops.com/tools/llm-security/", "category": "llm-security", "name": "AI Goat", "sourceUrl": "https://github.com/dhammon/ai-goat", "description": "vulnerable LLM CTF challenges", "directoryDescription": "AI Goat is included in the Awesome MLSecOps LLM Security and Red Teaming directory. The community-maintained README describes it as: “vulnerable LLM CTF challenges.” Its MLSecOps relevance is the testing or control of prompts, model behavior, retrieval paths, tool calls, outputs, or guardrails under adversarial input. The linked first-party source is the dhammon/ai-goat repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat and model coverage, Reproducible evaluations, CI and reporting support, and Sensitive-data handling. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "ai-scan-interceptor", "directoryUrl": "https://awesomemlsecops.com/tools/entries/ai-scan-interceptor/", "categoryUrl": "https://awesomemlsecops.com/tools/llm-security/", "category": "llm-security", "name": "AI-Scan-Interceptor", "sourceUrl": "https://github.com/mshirakawa-ssp/ai-scan-interceptor", "description": "Self-hostable DLP gateway for enterprise prompts to ChatGPT/Claude/Gemini (Squid + Go ICAP + mTLS, AGPL-3.0)", "directoryDescription": "AI-Scan-Interceptor is included in the Awesome MLSecOps LLM Security and Red Teaming directory. The community-maintained README describes it as: “Self-hostable DLP gateway for enterprise prompts to ChatGPT/Claude/Gemini (Squid + Go ICAP + mTLS, AGPL-3.0).” Its MLSecOps relevance is the testing or control of prompts, model behavior, retrieval paths, tool calls, outputs, or guardrails under adversarial input. The linked first-party source is the mshirakawa-ssp/ai-scan-interceptor repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat and model coverage, Reproducible evaluations, CI and reporting support, and Sensitive-data handling. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "artkit", "directoryUrl": "https://awesomemlsecops.com/tools/entries/artkit/", "categoryUrl": "https://awesomemlsecops.com/tools/llm-security/", "category": "llm-security", "name": "ARTkit", "sourceUrl": "https://github.com/BCG-X-Official/artkit", "description": "Automated prompt-based testing and evaluation of Gen AI applications", "directoryDescription": "ARTkit is included in the Awesome MLSecOps LLM Security and Red Teaming directory. The community-maintained README describes it as: “Automated prompt-based testing and evaluation of Gen AI applications.” Its MLSecOps relevance is the testing or control of prompts, model behavior, retrieval paths, tool calls, outputs, or guardrails under adversarial input. The linked first-party source is the BCG-X-Official/artkit repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat and model coverage, Reproducible evaluations, CI and reporting support, and Sensitive-data handling. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "damn-vulnerable-llm-project", "directoryUrl": "https://awesomemlsecops.com/tools/entries/damn-vulnerable-llm-project/", "categoryUrl": "https://awesomemlsecops.com/tools/llm-security/", "category": "llm-security", "name": "Damn Vulnerable LLM Project", "sourceUrl": "https://github.com/harishsg993010/DamnVulnerableLLMProject", "description": "A Large Language Model designed for getting hacked", "directoryDescription": "Damn Vulnerable LLM Project is included in the Awesome MLSecOps LLM Security and Red Teaming directory. The community-maintained README describes it as: “A Large Language Model designed for getting hacked.” Its MLSecOps relevance is the testing or control of prompts, model behavior, retrieval paths, tool calls, outputs, or guardrails under adversarial input. The linked first-party source is the harishsg993010/DamnVulnerableLLMProject repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat and model coverage, Reproducible evaluations, CI and reporting support, and Sensitive-data handling. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "gandalf-lakera", "directoryUrl": "https://awesomemlsecops.com/tools/entries/gandalf-lakera/", "categoryUrl": "https://awesomemlsecops.com/tools/llm-security/", "category": "llm-security", "name": "Gandalf Lakera", "sourceUrl": "https://gandalf.lakera.ai/", "description": "Prompt Injection CTF playground", "directoryDescription": "Gandalf Lakera is included in the Awesome MLSecOps LLM Security and Red Teaming directory. The community-maintained README describes it as: “Prompt Injection CTF playground.” Its MLSecOps relevance is the testing or control of prompts, model behavior, retrieval paths, tool calls, outputs, or guardrails under adversarial input. The linked first-party source is published at gandalf.lakera.ai. A technical review should test the project's documented evidence across four criteria: Threat and model coverage, Reproducible evaluations, CI and reporting support, and Sensitive-data handling. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "garak", "directoryUrl": "https://awesomemlsecops.com/tools/entries/garak/", "categoryUrl": "https://awesomemlsecops.com/tools/llm-security/", "category": "llm-security", "name": "Garak", "sourceUrl": "https://github.com/NVIDIA/garak", "description": "LLM vulnerability scanner", "directoryDescription": "Garak is included in the Awesome MLSecOps LLM Security and Red Teaming directory. The community-maintained README describes it as: “LLM vulnerability scanner.” Its MLSecOps relevance is the testing or control of prompts, model behavior, retrieval paths, tool calls, outputs, or guardrails under adversarial input. The linked first-party source is the NVIDIA/garak repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat and model coverage, Reproducible evaluations, CI and reporting support, and Sensitive-data handling. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "giskard", "directoryUrl": "https://awesomemlsecops.com/tools/entries/giskard/", "categoryUrl": "https://awesomemlsecops.com/tools/llm-security/", "category": "llm-security", "name": "Giskard", "sourceUrl": "https://github.com/Giskard-AI/giskard", "description": "Open-source testing tool for LLM applications", "directoryDescription": "Giskard is included in the Awesome MLSecOps LLM Security and Red Teaming directory. The community-maintained README describes it as: “Open-source testing tool for LLM applications.” Its MLSecOps relevance is the testing or control of prompts, model behavior, retrieval paths, tool calls, outputs, or guardrails under adversarial input. The linked first-party source is the Giskard-AI/giskard repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat and model coverage, Reproducible evaluations, CI and reporting support, and Sensitive-data handling. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "langbite", "directoryUrl": "https://awesomemlsecops.com/tools/entries/langbite/", "categoryUrl": "https://awesomemlsecops.com/tools/llm-security/", "category": "llm-security", "name": "LangBiTe", "sourceUrl": "https://github.com/SOM-Research/LangBiTe", "description": "A Bias Tester framework for LLMs", "directoryDescription": "LangBiTe is included in the Awesome MLSecOps LLM Security and Red Teaming directory. The community-maintained README describes it as: “A Bias Tester framework for LLMs.” Its MLSecOps relevance is the testing or control of prompts, model behavior, retrieval paths, tool calls, outputs, or guardrails under adversarial input. The linked first-party source is the SOM-Research/LangBiTe repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat and model coverage, Reproducible evaluations, CI and reporting support, and Sensitive-data handling. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "nemo-guardrails", "directoryUrl": "https://awesomemlsecops.com/tools/entries/nemo-guardrails/", "categoryUrl": "https://awesomemlsecops.com/tools/llm-security/", "category": "llm-security", "name": "NeMo Guardrails", "sourceUrl": "https://github.com/NVIDIA/NeMo-Guardrails", "description": "NeMo Guardrails allow developers building LLM-based applications to add programmable guardrails between the application code and the LLM", "directoryDescription": "NeMo Guardrails is included in the Awesome MLSecOps LLM Security and Red Teaming directory. The community-maintained README describes it as: “NeMo Guardrails allow developers building LLM-based applications to add programmable guardrails between the application code and the LLM.” Its MLSecOps relevance is the testing or control of prompts, model behavior, retrieval paths, tool calls, outputs, or guardrails under adversarial input. The linked first-party source is the NVIDIA/NeMo-Guardrails repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat and model coverage, Reproducible evaluations, CI and reporting support, and Sensitive-data handling. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "neuraltrust", "directoryUrl": "https://awesomemlsecops.com/tools/entries/neuraltrust/", "categoryUrl": "https://awesomemlsecops.com/tools/llm-security/", "category": "llm-security", "name": "NeuralTrust", "sourceUrl": "https://neuraltrust.ai/", "description": "Tools to protect, secure and test GenAI Applications", "directoryDescription": "NeuralTrust is included in the Awesome MLSecOps LLM Security and Red Teaming directory. The community-maintained README describes it as: “Tools to protect, secure and test GenAI Applications.” Its MLSecOps relevance is the testing or control of prompts, model behavior, retrieval paths, tool calls, outputs, or guardrails under adversarial input. The linked first-party source is published at neuraltrust.ai. A technical review should test the project's documented evidence across four criteria: Threat and model coverage, Reproducible evaluations, CI and reporting support, and Sensitive-data handling. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "pallms-payloads-for-attacking-large-language-models", "directoryUrl": "https://awesomemlsecops.com/tools/entries/pallms-payloads-for-attacking-large-language-models/", "categoryUrl": "https://awesomemlsecops.com/tools/llm-security/", "category": "llm-security", "name": "PALLMs (Payloads for Attacking Large Language Models)", "sourceUrl": "https://github.com/mik0w/pallms", "description": "list of various payloads for attacking LLMs collected in one place", "directoryDescription": "PALLMs (Payloads for Attacking Large Language Models) is included in the Awesome MLSecOps LLM Security and Red Teaming directory. The community-maintained README describes it as: “list of various payloads for attacking LLMs collected in one place.” Its MLSecOps relevance is the testing or control of prompts, model behavior, retrieval paths, tool calls, outputs, or guardrails under adversarial input. The linked first-party source is the mik0w/pallms repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat and model coverage, Reproducible evaluations, CI and reporting support, and Sensitive-data handling. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "promptfoo-scanner", "directoryUrl": "https://awesomemlsecops.com/tools/entries/promptfoo-scanner/", "categoryUrl": "https://awesomemlsecops.com/tools/llm-security/", "category": "llm-security", "name": "Promptfoo Scanner", "sourceUrl": "https://github.com/promptfoo/promptfoo", "description": "An open-source LLM red teaming tool", "directoryDescription": "Promptfoo Scanner is included in the Awesome MLSecOps LLM Security and Red Teaming directory. The community-maintained README describes it as: “An open-source LLM red teaming tool.” Its MLSecOps relevance is the testing or control of prompts, model behavior, retrieval paths, tool calls, outputs, or guardrails under adversarial input. The linked first-party source is the promptfoo/promptfoo repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat and model coverage, Reproducible evaluations, CI and reporting support, and Sensitive-data handling. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "promptinject", "directoryUrl": "https://awesomemlsecops.com/tools/entries/promptinject/", "categoryUrl": "https://awesomemlsecops.com/tools/llm-security/", "category": "llm-security", "name": "PromptInject", "sourceUrl": "https://github.com/agencyenterprise/PromptInject", "description": "A framework that assembles adversarial prompts", "directoryDescription": "PromptInject is included in the Awesome MLSecOps LLM Security and Red Teaming directory. The community-maintained README describes it as: “A framework that assembles adversarial prompts.” Its MLSecOps relevance is the testing or control of prompts, model behavior, retrieval paths, tool calls, outputs, or guardrails under adversarial input. The linked first-party source is the agencyenterprise/PromptInject repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat and model coverage, Reproducible evaluations, CI and reporting support, and Sensitive-data handling. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "ps-fuzz", "directoryUrl": "https://awesomemlsecops.com/tools/entries/ps-fuzz/", "categoryUrl": "https://awesomemlsecops.com/tools/llm-security/", "category": "llm-security", "name": "PS-fuzz", "sourceUrl": "https://github.com/prompt-security/ps-fuzz", "description": "tool for scanning LLM vulnerabilities", "directoryDescription": "PS-fuzz is included in the Awesome MLSecOps LLM Security and Red Teaming directory. The community-maintained README describes it as: “tool for scanning LLM vulnerabilities.” Its MLSecOps relevance is the testing or control of prompts, model behavior, retrieval paths, tool calls, outputs, or guardrails under adversarial input. The linked first-party source is the prompt-security/ps-fuzz repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat and model coverage, Reproducible evaluations, CI and reporting support, and Sensitive-data handling. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "purple-llama", "directoryUrl": "https://awesomemlsecops.com/tools/entries/purple-llama/", "categoryUrl": "https://awesomemlsecops.com/tools/llm-security/", "category": "llm-security", "name": "Purple Llama", "sourceUrl": "https://meta-llama.github.io/PurpleLlama/", "description": "Meta's umbrella suite of LLM safety and security tooling, including Llama Guard, Code Shield, and CyberSecEval", "directoryDescription": "Purple Llama is included in the Awesome MLSecOps LLM Security and Red Teaming directory. The community-maintained README describes it as: “Meta's umbrella suite of LLM safety and security tooling, including Llama Guard, Code Shield, and CyberSecEval.” Its MLSecOps relevance is the testing or control of prompts, model behavior, retrieval paths, tool calls, outputs, or guardrails under adversarial input. The linked first-party source is published at meta-llama.github.io. A technical review should test the project's documented evidence across four criteria: Threat and model coverage, Reproducible evaluations, CI and reporting support, and Sensitive-data handling. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "pyrit", "directoryUrl": "https://awesomemlsecops.com/tools/entries/pyrit/", "categoryUrl": "https://awesomemlsecops.com/tools/llm-security/", "category": "llm-security", "name": "Pyrit", "sourceUrl": "https://github.com/Azure/PyRIT", "description": "The Python Risk Identification Tool for generative AI", "directoryDescription": "Pyrit is included in the Awesome MLSecOps LLM Security and Red Teaming directory. The community-maintained README describes it as: “The Python Risk Identification Tool for generative AI.” Its MLSecOps relevance is the testing or control of prompts, model behavior, retrieval paths, tool calls, outputs, or guardrails under adversarial input. The linked first-party source is the Azure/PyRIT repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat and model coverage, Reproducible evaluations, CI and reporting support, and Sensitive-data handling. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "trustgate", "directoryUrl": "https://awesomemlsecops.com/tools/entries/trustgate/", "categoryUrl": "https://awesomemlsecops.com/tools/llm-security/", "category": "llm-security", "name": "TrustGate", "sourceUrl": "https://github.com/NeuralTrust/TrustGate", "description": "An open-source Generative Application Firewall (GAF)", "directoryDescription": "TrustGate is included in the Awesome MLSecOps LLM Security and Red Teaming directory. The community-maintained README describes it as: “An open-source Generative Application Firewall (GAF).” Its MLSecOps relevance is the testing or control of prompts, model behavior, retrieval paths, tool calls, outputs, or guardrails under adversarial input. The linked first-party source is the NeuralTrust/TrustGate repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat and model coverage, Reproducible evaluations, CI and reporting support, and Sensitive-data handling. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "vigil", "directoryUrl": "https://awesomemlsecops.com/tools/entries/vigil/", "categoryUrl": "https://awesomemlsecops.com/tools/llm-security/", "category": "llm-security", "name": "Vigil", "sourceUrl": "https://github.com/deadbits/vigil-llm", "description": "LLM prompt injection and security scanner", "directoryDescription": "Vigil is included in the Awesome MLSecOps LLM Security and Red Teaming directory. The community-maintained README describes it as: “LLM prompt injection and security scanner.” Its MLSecOps relevance is the testing or control of prompts, model behavior, retrieval paths, tool calls, outputs, or guardrails under adversarial input. The linked first-party source is the deadbits/vigil-llm repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat and model coverage, Reproducible evaluations, CI and reporting support, and Sensitive-data handling. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "whistleblower", "directoryUrl": "https://awesomemlsecops.com/tools/entries/whistleblower/", "categoryUrl": "https://awesomemlsecops.com/tools/llm-security/", "category": "llm-security", "name": "Whistleblower", "sourceUrl": "https://github.com/Repello-AI/whistleblower", "description": "Open-source offensive tool by Repello AI for testing LLM apps against system prompt leakage", "directoryDescription": "Whistleblower is included in the Awesome MLSecOps LLM Security and Red Teaming directory. The community-maintained README describes it as: “Open-source offensive tool by Repello AI for testing LLM apps against system prompt leakage.” Its MLSecOps relevance is the testing or control of prompts, model behavior, retrieval paths, tool calls, outputs, or guardrails under adversarial input. The linked first-party source is the Repello-AI/whistleblower repository on GitHub. A technical review should test the project's documented evidence across four criteria: Threat and model coverage, Reproducible evaluations, CI and reporting support, and Sensitive-data handling. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "aishield-watchtower", "directoryUrl": "https://awesomemlsecops.com/tools/entries/aishield-watchtower/", "categoryUrl": "https://awesomemlsecops.com/tools/model-scanning/", "category": "model-scanning", "name": "AIShield Watchtower", "sourceUrl": "https://github.com/bosch-aisecurity-aishield/watchtower", "description": "An open-source tool from AIShield for studying AI models and scanning for vulnerabilities", "directoryDescription": "AIShield Watchtower is included in the Awesome MLSecOps Model Scanning and Validation directory. The community-maintained README describes it as: “An open-source tool from AIShield for studying AI models and scanning for vulnerabilities.” Its MLSecOps relevance is the inspection or validation of model artifacts, notebooks, code, dependencies, or model behavior before release and deployment. The linked first-party source is the bosch-aisecurity-aishield/watchtower repository on GitHub. A technical review should test the project's documented evidence across four criteria: Supported artifact formats, Detection evidence, False-positive handling, and CI and SARIF output. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "audit-ai", "directoryUrl": "https://awesomemlsecops.com/tools/entries/audit-ai/", "categoryUrl": "https://awesomemlsecops.com/tools/model-scanning/", "category": "model-scanning", "name": "Audit AI", "sourceUrl": "https://github.com/pymetrics/audit-ai", "description": "Bias Testing for Generalized Machine Learning Applications", "directoryDescription": "Audit AI is included in the Awesome MLSecOps Model Scanning and Validation directory. The community-maintained README describes it as: “Bias Testing for Generalized Machine Learning Applications.” Its MLSecOps relevance is the inspection or validation of model artifacts, notebooks, code, dependencies, or model behavior before release and deployment. The linked first-party source is the pymetrics/audit-ai repository on GitHub. A technical review should test the project's documented evidence across four criteria: Supported artifact formats, Detection evidence, False-positive handling, and CI and SARIF output. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "circleguardbench", "directoryUrl": "https://awesomemlsecops.com/tools/entries/circleguardbench/", "categoryUrl": "https://awesomemlsecops.com/tools/model-scanning/", "category": "model-scanning", "name": "CircleGuardBench", "sourceUrl": "https://github.com/whitecircle/circle-guard-bench", "description": "A full-fledged benchmark for evaluating protection capabilities of AI models", "directoryDescription": "CircleGuardBench is included in the Awesome MLSecOps Model Scanning and Validation directory. The community-maintained README describes it as: “A full-fledged benchmark for evaluating protection capabilities of AI models.” Its MLSecOps relevance is the inspection or validation of model artifacts, notebooks, code, dependencies, or model behavior before release and deployment. The linked first-party source is the whitecircle/circle-guard-bench repository on GitHub. A technical review should test the project's documented evidence across four criteria: Supported artifact formats, Detection evidence, False-positive handling, and CI and SARIF output. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "cisco-ai-defense-formerly-robust-intelligence", "directoryUrl": "https://awesomemlsecops.com/tools/entries/cisco-ai-defense-formerly-robust-intelligence/", "categoryUrl": "https://awesomemlsecops.com/tools/model-scanning/", "category": "model-scanning", "name": "Cisco AI Defense (formerly Robust Intelligence)", "sourceUrl": "https://www.cisco.com/site/us/en/products/security/ai-defense/index.html", "description": "Commercial platform for AI model validation and runtime protection; Robust Intelligence was acquired by Cisco", "directoryDescription": "Cisco AI Defense (formerly Robust Intelligence) is included in the Awesome MLSecOps Model Scanning and Validation directory. The community-maintained README describes it as: “Commercial platform for AI model validation and runtime protection; Robust Intelligence was acquired by Cisco.” Its MLSecOps relevance is the inspection or validation of model artifacts, notebooks, code, dependencies, or model behavior before release and deployment. The linked first-party source is published at cisco.com. A technical review should test the project's documented evidence across four criteria: Supported artifact formats, Detection evidence, False-positive handling, and CI and SARIF output. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "citadel-lens", "directoryUrl": "https://awesomemlsecops.com/tools/entries/citadel-lens/", "categoryUrl": "https://awesomemlsecops.com/tools/model-scanning/", "category": "model-scanning", "name": "Citadel Lens", "sourceUrl": "https://citadel-ai.com/products/lens/", "description": "Commercial model-testing platform; verify supported standards and security checks in current product documentation", "directoryDescription": "Citadel Lens is included in the Awesome MLSecOps Model Scanning and Validation directory. The community-maintained README describes it as: “Commercial model-testing platform; verify supported standards and security checks in current product documentation.” Its MLSecOps relevance is the inspection or validation of model artifacts, notebooks, code, dependencies, or model behavior before release and deployment. The linked first-party source is published at citadel-ai.com. A technical review should test the project's documented evidence across four criteria: Supported artifact formats, Detection evidence, False-positive handling, and CI and SARIF output. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "flawed-machine-learning-security", "directoryUrl": "https://awesomemlsecops.com/tools/entries/flawed-machine-learning-security/", "categoryUrl": "https://awesomemlsecops.com/tools/model-scanning/", "category": "model-scanning", "name": "Flawed Machine Learning Security", "sourceUrl": "https://github.com/EthicalML/fml-security", "description": "Practical examples of \"Flawed Machine Learning Security\" together with ML Security best practice across the end to end stages of the machine learning model lifecycle from training, to packaging, to deployment", "directoryDescription": "Flawed Machine Learning Security is included in the Awesome MLSecOps Model Scanning and Validation directory. The community-maintained README describes it as: “Practical examples of \"Flawed Machine Learning Security\" together with ML Security best practice across the end to end stages of the machine learning model lifecycle from training, to packaging, to deployment.” Its MLSecOps relevance is the inspection or validation of model artifacts, notebooks, code, dependencies, or model behavior before release and deployment. The linked first-party source is the EthicalML/fml-security repository on GitHub. A technical review should test the project's documented evidence across four criteria: Supported artifact formats, Detection evidence, False-positive handling, and CI and SARIF output. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "guardian", "directoryUrl": "https://awesomemlsecops.com/tools/entries/guardian/", "categoryUrl": "https://awesomemlsecops.com/tools/model-scanning/", "category": "model-scanning", "name": "Guardian", "sourceUrl": "https://protectai.com/guardian", "description": "Model protection in CI/CD", "directoryDescription": "Guardian is included in the Awesome MLSecOps Model Scanning and Validation directory. The community-maintained README describes it as: “Model protection in CI/CD.” Its MLSecOps relevance is the inspection or validation of model artifacts, notebooks, code, dependencies, or model behavior before release and deployment. The linked first-party source is published at protectai.com. A technical review should test the project's documented evidence across four criteria: Supported artifact formats, Detection evidence, False-positive handling, and CI and SARIF output. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "hiddenlayer-ai-detection-and-response", "directoryUrl": "https://awesomemlsecops.com/tools/entries/hiddenlayer-ai-detection-and-response/", "categoryUrl": "https://awesomemlsecops.com/tools/model-scanning/", "category": "model-scanning", "name": "HiddenLayer AI Detection and Response", "sourceUrl": "https://www.hiddenlayer.com/platform", "description": "Commercial AI detection and response platform from HiddenLayer", "directoryDescription": "HiddenLayer AI Detection and Response is included in the Awesome MLSecOps Model Scanning and Validation directory. The community-maintained README describes it as: “Commercial AI detection and response platform from HiddenLayer.” Its MLSecOps relevance is the inspection or validation of model artifacts, notebooks, code, dependencies, or model behavior before release and deployment. The linked first-party source is published at hiddenlayer.com. A technical review should test the project's documented evidence across four criteria: Supported artifact formats, Detection evidence, False-positive handling, and CI and SARIF output. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "hiddenlayer-model-as-code", "directoryUrl": "https://awesomemlsecops.com/tools/entries/hiddenlayer-model-as-code/", "categoryUrl": "https://awesomemlsecops.com/tools/model-scanning/", "category": "model-scanning", "name": "HiddenLayer: Model as Code", "sourceUrl": "https://hiddenlayer.com/research/models-are-code/", "description": "Research article on executable model formats and unsafe deserialization paths in ML libraries", "directoryDescription": "HiddenLayer: Model as Code is included in the Awesome MLSecOps Model Scanning and Validation directory. The community-maintained README describes it as: “Research article on executable model formats and unsafe deserialization paths in ML libraries.” Its MLSecOps relevance is the inspection or validation of model artifacts, notebooks, code, dependencies, or model behavior before release and deployment. The linked first-party source is published at hiddenlayer.com. A technical review should test the project's documented evidence across four criteria: Supported artifact formats, Detection evidence, False-positive handling, and CI and SARIF output. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "lintml", "directoryUrl": "https://awesomemlsecops.com/tools/entries/lintml/", "categoryUrl": "https://awesomemlsecops.com/tools/model-scanning/", "category": "model-scanning", "name": "lintML", "sourceUrl": "https://github.com/JosephTLucas/lintML", "description": "Security linter for ML training code, maintained by a member of the NVIDIA AI red team", "directoryDescription": "lintML is included in the Awesome MLSecOps Model Scanning and Validation directory. The community-maintained README describes it as: “Security linter for ML training code, maintained by a member of the NVIDIA AI red team.” Its MLSecOps relevance is the inspection or validation of model artifacts, notebooks, code, dependencies, or model behavior before release and deployment. The linked first-party source is the JosephTLucas/lintML repository on GitHub. A technical review should test the project's documented evidence across four criteria: Supported artifact formats, Detection evidence, False-positive handling, and CI and SARIF output. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "mindgard-cli", "directoryUrl": "https://awesomemlsecops.com/tools/entries/mindgard-cli/", "categoryUrl": "https://awesomemlsecops.com/tools/model-scanning/", "category": "model-scanning", "name": "Mindgard CLI", "sourceUrl": "https://github.com/Mindgard/cli/", "description": "Evaluate the security of AI systems through a CLI", "directoryDescription": "Mindgard CLI is included in the Awesome MLSecOps Model Scanning and Validation directory. The community-maintained README describes it as: “Evaluate the security of AI systems through a CLI.” Its MLSecOps relevance is the inspection or validation of model artifacts, notebooks, code, dependencies, or model behavior before release and deployment. The linked first-party source is the Mindgard/cli repository on GitHub. A technical review should test the project's documented evidence across four criteria: Supported artifact formats, Detection evidence, False-positive handling, and CI and SARIF output. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "nb-defense", "directoryUrl": "https://awesomemlsecops.com/tools/entries/nb-defense/", "categoryUrl": "https://awesomemlsecops.com/tools/model-scanning/", "category": "model-scanning", "name": "NB Defense", "sourceUrl": "https://github.com/protectai/nbdefense", "description": "Jupyter notebook security scanner from Protect AI (now part of Palo Alto Networks)", "directoryDescription": "NB Defense is included in the Awesome MLSecOps Model Scanning and Validation directory. The community-maintained README describes it as: “Jupyter notebook security scanner from Protect AI (now part of Palo Alto Networks).” Its MLSecOps relevance is the inspection or validation of model artifacts, notebooks, code, dependencies, or model behavior before release and deployment. The linked first-party source is the protectai/nbdefense repository on GitHub. A technical review should test the project's documented evidence across four criteria: Supported artifact formats, Detection evidence, False-positive handling, and CI and SARIF output. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "tensorflow-model-analysis", "directoryUrl": "https://awesomemlsecops.com/tools/entries/tensorflow-model-analysis/", "categoryUrl": "https://awesomemlsecops.com/tools/model-scanning/", "category": "model-scanning", "name": "TensorFlow Model Analysis", "sourceUrl": "https://github.com/tensorflow/model-analysis", "description": "A library for analyzing, validating, and monitoring machine learning models in production", "directoryDescription": "TensorFlow Model Analysis is included in the Awesome MLSecOps Model Scanning and Validation directory. The community-maintained README describes it as: “A library for analyzing, validating, and monitoring machine learning models in production.” Its MLSecOps relevance is the inspection or validation of model artifacts, notebooks, code, dependencies, or model behavior before release and deployment. The linked first-party source is the tensorflow/model-analysis repository on GitHub. A technical review should test the project's documented evidence across four criteria: Supported artifact formats, Detection evidence, False-positive handling, and CI and SARIF output. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "anonllm", "directoryUrl": "https://awesomemlsecops.com/tools/entries/anonllm/", "categoryUrl": "https://awesomemlsecops.com/tools/privacy/", "category": "privacy", "name": "AnonLLM", "sourceUrl": "https://github.com/fsndzomga/anonLLM", "description": "Anonymize Personally Identifiable Information (PII) for Large Language Model APIs", "directoryDescription": "AnonLLM is included in the Awesome MLSecOps Privacy-Preserving Machine Learning directory. The community-maintained README describes it as: “Anonymize Personally Identifiable Information (PII) for Large Language Model APIs.” Its MLSecOps relevance is the reduction or measurement of sensitive-data exposure through differential privacy, anonymization, encrypted computation, or privacy attack testing. The linked first-party source is the fsndzomga/anonLLM repository on GitHub. A technical review should test the project's documented evidence across four criteria: Explicit privacy guarantees, Utility and accuracy impact, Re-identification testing, and Repeatable privacy accounting. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "arx-data-anonymization-tool", "directoryUrl": "https://awesomemlsecops.com/tools/entries/arx-data-anonymization-tool/", "categoryUrl": "https://awesomemlsecops.com/tools/privacy/", "category": "privacy", "name": "ARX - Data Anonymization Tool", "sourceUrl": "https://arx.deidentifier.org/", "description": "Tool for anonymizing datasets", "directoryDescription": "ARX - Data Anonymization Tool is included in the Awesome MLSecOps Privacy-Preserving Machine Learning directory. The community-maintained README describes it as: “Tool for anonymizing datasets.” Its MLSecOps relevance is the reduction or measurement of sensitive-data exposure through differential privacy, anonymization, encrypted computation, or privacy attack testing. The linked first-party source is published at arx.deidentifier.org. A technical review should test the project's documented evidence across four criteria: Explicit privacy guarantees, Utility and accuracy impact, Re-identification testing, and Repeatable privacy accounting. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "bmw-anonymization-api", "directoryUrl": "https://awesomemlsecops.com/tools/entries/bmw-anonymization-api/", "categoryUrl": "https://awesomemlsecops.com/tools/privacy/", "category": "privacy", "name": "BMW Anonymization API", "sourceUrl": "https://github.com/BMW-InnovationLab/BMW-Anonymization-API", "description": "This repository allows you to anonymize sensitive information in images/videos. The solution is fully compatible with the DL-based training/inference solutions that we already published/will publish for Object Detection and Semantic Segmentation", "directoryDescription": "BMW Anonymization API is included in the Awesome MLSecOps Privacy-Preserving Machine Learning directory. The community-maintained README describes it as: “This repository allows you to anonymize sensitive information in images/videos. The solution is fully compatible with the DL-based training/inference solutions that we already published/will publish for Object Detection and Semantic Segmentation.” Its MLSecOps relevance is the reduction or measurement of sensitive-data exposure through differential privacy, anonymization, encrypted computation, or privacy attack testing. The linked first-party source is the BMW-InnovationLab/BMW-Anonymization-API repository on GitHub. A technical review should test the project's documented evidence across four criteria: Explicit privacy guarantees, Utility and accuracy impact, Re-identification testing, and Repeatable privacy accounting. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "data-veil", "directoryUrl": "https://awesomemlsecops.com/tools/entries/data-veil/", "categoryUrl": "https://awesomemlsecops.com/tools/privacy/", "category": "privacy", "name": "Data-Veil", "sourceUrl": "https://veil.ai/", "description": "Data masking and anonymization tool", "directoryDescription": "Data-Veil is included in the Awesome MLSecOps Privacy-Preserving Machine Learning directory. The community-maintained README describes it as: “Data masking and anonymization tool.” Its MLSecOps relevance is the reduction or measurement of sensitive-data exposure through differential privacy, anonymization, encrypted computation, or privacy attack testing. The linked first-party source is published at veil.ai. A technical review should test the project's documented evidence across four criteria: Explicit privacy guarantees, Utility and accuracy impact, Re-identification testing, and Repeatable privacy accounting. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "deepprivacy2", "directoryUrl": "https://awesomemlsecops.com/tools/entries/deepprivacy2/", "categoryUrl": "https://awesomemlsecops.com/tools/privacy/", "category": "privacy", "name": "DeepPrivacy2", "sourceUrl": "https://github.com/hukkelas/deep_privacy2", "description": "A Toolbox for Realistic Image Anonymization", "directoryDescription": "DeepPrivacy2 is included in the Awesome MLSecOps Privacy-Preserving Machine Learning directory. The community-maintained README describes it as: “A Toolbox for Realistic Image Anonymization.” Its MLSecOps relevance is the reduction or measurement of sensitive-data exposure through differential privacy, anonymization, encrypted computation, or privacy attack testing. The linked first-party source is the hukkelas/deep_privacy2 repository on GitHub. A technical review should test the project's documented evidence across four criteria: Explicit privacy guarantees, Utility and accuracy impact, Re-identification testing, and Repeatable privacy accounting. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "differential-privacy-library", "directoryUrl": "https://awesomemlsecops.com/tools/entries/differential-privacy-library/", "categoryUrl": "https://awesomemlsecops.com/tools/privacy/", "category": "privacy", "name": "differential-privacy-library", "sourceUrl": "https://github.com/IBM/differential-privacy-library", "description": "Library designed for differential privacy and machine learning", "directoryDescription": "differential-privacy-library is included in the Awesome MLSecOps Privacy-Preserving Machine Learning directory. The community-maintained README describes it as: “Library designed for differential privacy and machine learning.” Its MLSecOps relevance is the reduction or measurement of sensitive-data exposure through differential privacy, anonymization, encrypted computation, or privacy attack testing. The linked first-party source is the IBM/differential-privacy-library repository on GitHub. A technical review should test the project's documented evidence across four criteria: Explicit privacy guarantees, Utility and accuracy impact, Re-identification testing, and Repeatable privacy accounting. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "opendp", "directoryUrl": "https://awesomemlsecops.com/tools/entries/opendp/", "categoryUrl": "https://awesomemlsecops.com/tools/privacy/", "category": "privacy", "name": "OpenDP", "sourceUrl": "https://github.com/opendp/opendp", "description": "The core library of differential privacy algorithms powering the OpenDP Project", "directoryDescription": "OpenDP is included in the Awesome MLSecOps Privacy-Preserving Machine Learning directory. The community-maintained README describes it as: “The core library of differential privacy algorithms powering the OpenDP Project.” Its MLSecOps relevance is the reduction or measurement of sensitive-data exposure through differential privacy, anonymization, encrypted computation, or privacy attack testing. The linked first-party source is the opendp/opendp repository on GitHub. A technical review should test the project's documented evidence across four criteria: Explicit privacy guarantees, Utility and accuracy impact, Re-identification testing, and Repeatable privacy accounting. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "privacy-meter", "directoryUrl": "https://awesomemlsecops.com/tools/entries/privacy-meter/", "categoryUrl": "https://awesomemlsecops.com/tools/privacy/", "category": "privacy", "name": "Privacy Meter", "sourceUrl": "https://github.com/privacytrustlab/ml_privacy_meter", "description": "An open-source library to audit data privacy in statistical and machine learning algorithms", "directoryDescription": "Privacy Meter is included in the Awesome MLSecOps Privacy-Preserving Machine Learning directory. The community-maintained README describes it as: “An open-source library to audit data privacy in statistical and machine learning algorithms.” Its MLSecOps relevance is the reduction or measurement of sensitive-data exposure through differential privacy, anonymization, encrypted computation, or privacy attack testing. The linked first-party source is the privacytrustlab/ml_privacy_meter repository on GitHub. A technical review should test the project's documented evidence across four criteria: Explicit privacy guarantees, Utility and accuracy impact, Re-identification testing, and Repeatable privacy accounting. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "tensorflow-privacy", "directoryUrl": "https://awesomemlsecops.com/tools/entries/tensorflow-privacy/", "categoryUrl": "https://awesomemlsecops.com/tools/privacy/", "category": "privacy", "name": "TensorFlow Privacy", "sourceUrl": "https://github.com/tensorflow/privacy", "description": "Library of privacy-preserving machine learning algorithms and tools", "directoryDescription": "TensorFlow Privacy is included in the Awesome MLSecOps Privacy-Preserving Machine Learning directory. The community-maintained README describes it as: “Library of privacy-preserving machine learning algorithms and tools.” Its MLSecOps relevance is the reduction or measurement of sensitive-data exposure through differential privacy, anonymization, encrypted computation, or privacy attack testing. The linked first-party source is the tensorflow/privacy repository on GitHub. A technical review should test the project's documented evidence across four criteria: Explicit privacy guarantees, Utility and accuracy impact, Re-identification testing, and Repeatable privacy accounting. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "tf-encrypted", "directoryUrl": "https://awesomemlsecops.com/tools/entries/tf-encrypted/", "categoryUrl": "https://awesomemlsecops.com/tools/privacy/", "category": "privacy", "name": "TF Encrypted", "sourceUrl": "https://github.com/tf-encrypted/tf-encrypted", "description": "Framework for encrypted deep learning in TensorFlow using secure multi-party computation", "directoryDescription": "TF Encrypted is included in the Awesome MLSecOps Privacy-Preserving Machine Learning directory. The community-maintained README describes it as: “Framework for encrypted deep learning in TensorFlow using secure multi-party computation.” Its MLSecOps relevance is the reduction or measurement of sensitive-data exposure through differential privacy, anonymization, encrypted computation, or privacy attack testing. The linked first-party source is the tf-encrypted/tf-encrypted repository on GitHub. A technical review should test the project's documented evidence across four criteria: Explicit privacy guarantees, Utility and accuracy impact, Re-identification testing, and Repeatable privacy accounting. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "tool-for-data-anonymization", "directoryUrl": "https://awesomemlsecops.com/tools/entries/tool-for-data-anonymization/", "categoryUrl": "https://awesomemlsecops.com/tools/privacy/", "category": "privacy", "name": "Tool for DATA anonymization", "sourceUrl": "https://github.com/PacktPublishing/Adversarial-AI---Attacks-Mitigations-and-Defense-Strategies/blob/main/ch10/notebooks/Data%20Anonymization.ipynb", "description": "Data anonymization", "directoryDescription": "Tool for DATA anonymization is included in the Awesome MLSecOps Privacy-Preserving Machine Learning directory. The community-maintained README describes it as: “Data anonymization.” Its MLSecOps relevance is the reduction or measurement of sensitive-data exposure through differential privacy, anonymization, encrypted computation, or privacy attack testing. The linked first-party source is the PacktPublishing/Adversarial-AI---Attacks-Mitigations-and-Defense-Strategies repository on GitHub. A technical review should test the project's documented evidence across four criteria: Explicit privacy guarantees, Utility and accuracy impact, Re-identification testing, and Repeatable privacy accounting. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "tool-for-img-anonymization", "directoryUrl": "https://awesomemlsecops.com/tools/entries/tool-for-img-anonymization/", "categoryUrl": "https://awesomemlsecops.com/tools/privacy/", "category": "privacy", "name": "Tool for IMG anonymization", "sourceUrl": "https://github.com/PacktPublishing/Adversarial-AI---Attacks-Mitigations-and-Defense-Strategies/blob/main/ch10/notebooks/Image%20Anonymization.ipynb", "description": "Image anonymization", "directoryDescription": "Tool for IMG anonymization is included in the Awesome MLSecOps Privacy-Preserving Machine Learning directory. The community-maintained README describes it as: “Image anonymization.” Its MLSecOps relevance is the reduction or measurement of sensitive-data exposure through differential privacy, anonymization, encrypted computation, or privacy attack testing. The linked first-party source is the PacktPublishing/Adversarial-AI---Attacks-Mitigations-and-Defense-Strategies repository on GitHub. A technical review should test the project's documented evidence across four criteria: Explicit privacy guarantees, Utility and accuracy impact, Re-identification testing, and Repeatable privacy accounting. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "ai-exploits", "directoryUrl": "https://awesomemlsecops.com/tools/entries/ai-exploits/", "categoryUrl": "https://awesomemlsecops.com/tools/supply-chain/", "category": "supply-chain", "name": "AI-exploits", "sourceUrl": "https://github.com/protectai/ai-exploits", "description": "Exploits for MLOps systems, extending beyond inputs provided to LLMs such as ChatGPT", "directoryDescription": "AI-exploits is included in the Awesome MLSecOps AI Supply-Chain Security directory. The community-maintained README describes it as: “Exploits for MLOps systems, extending beyond inputs provided to LLMs such as ChatGPT.” Its MLSecOps relevance is the protection of model provenance, artifact integrity, dependencies, signing, bills of materials, registries, or delivery pipelines. The linked first-party source is the protectai/ai-exploits repository on GitHub. A technical review should test the project's documented evidence across four criteria: Provenance and signing support, ML-BOM formats, Registry and CI integration, and Policy enforcement. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "aisbom", "directoryUrl": "https://awesomemlsecops.com/tools/entries/aisbom/", "categoryUrl": "https://awesomemlsecops.com/tools/supply-chain/", "category": "supply-chain", "name": "AIsbom", "sourceUrl": "https://github.com/Lab700xOrg/aisbom", "description": "Disassembles Pickle bytecode and parses SafeTensors/GGUF binary headers to detect malware and license risks in ML model files before load. Generates CycloneDX/SPDX SBOMs.", "directoryDescription": "AIsbom is included in the Awesome MLSecOps AI Supply-Chain Security directory. The community-maintained README describes it as: “Disassembles Pickle bytecode and parses SafeTensors/GGUF binary headers to detect malware and license risks in ML model files before load. Generates CycloneDX/SPDX SBOMs.” Its MLSecOps relevance is the protection of model provenance, artifact integrity, dependencies, signing, bills of materials, registries, or delivery pipelines. The linked first-party source is the Lab700xOrg/aisbom repository on GitHub. A technical review should test the project's documented evidence across four criteria: Provenance and signing support, ML-BOM formats, Registry and CI integration, and Policy enforcement. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "bomlens", "directoryUrl": "https://awesomemlsecops.com/tools/entries/bomlens/", "categoryUrl": "https://awesomemlsecops.com/tools/supply-chain/", "category": "supply-chain", "name": "BomLens", "sourceUrl": "https://github.com/sktelecom/bomlens", "description": "Local-first SBOM generator that builds CycloneDX ML-BOMs for Hugging Face models, with license and known-vulnerability reports", "directoryDescription": "BomLens is included in the Awesome MLSecOps AI Supply-Chain Security directory. The community-maintained README describes it as: “Local-first SBOM generator that builds CycloneDX ML-BOMs for Hugging Face models, with license and known-vulnerability reports.” Its MLSecOps relevance is the protection of model provenance, artifact integrity, dependencies, signing, bills of materials, registries, or delivery pipelines. The linked first-party source is the sktelecom/bomlens repository on GitHub. A technical review should test the project's documented evidence across four criteria: Provenance and signing support, ML-BOM formats, Registry and CI integration, and Policy enforcement. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "model-transparency", "directoryUrl": "https://awesomemlsecops.com/tools/entries/model-transparency/", "categoryUrl": "https://awesomemlsecops.com/tools/supply-chain/", "category": "supply-chain", "name": "Model Transparency", "sourceUrl": "https://github.com/sigstore/model-transparency", "description": "Generate model signing metadata for provenance verification", "directoryDescription": "Model Transparency is included in the Awesome MLSecOps AI Supply-Chain Security directory. The community-maintained README describes it as: “Generate model signing metadata for provenance verification.” Its MLSecOps relevance is the protection of model provenance, artifact integrity, dependencies, signing, bills of materials, registries, or delivery pipelines. The linked first-party source is the sigstore/model-transparency repository on GitHub. A technical review should test the project's documented evidence across four criteria: Provenance and signing support, ML-BOM formats, Registry and CI integration, and Policy enforcement. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "modelscan", "directoryUrl": "https://awesomemlsecops.com/tools/entries/modelscan/", "categoryUrl": "https://awesomemlsecops.com/tools/supply-chain/", "category": "supply-chain", "name": "ModelScan", "sourceUrl": "https://github.com/protectai/modelscan", "description": "Protection Against ML Model Serialization Attacks", "directoryDescription": "ModelScan is included in the Awesome MLSecOps AI Supply-Chain Security directory. The community-maintained README describes it as: “Protection Against ML Model Serialization Attacks.” Its MLSecOps relevance is the protection of model provenance, artifact integrity, dependencies, signing, bills of materials, registries, or delivery pipelines. The linked first-party source is the protectai/modelscan repository on GitHub. A technical review should test the project's documented evidence across four criteria: Provenance and signing support, ML-BOM formats, Registry and CI integration, and Policy enforcement. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" }, { "id": "safetensors", "directoryUrl": "https://awesomemlsecops.com/tools/entries/safetensors/", "categoryUrl": "https://awesomemlsecops.com/tools/supply-chain/", "category": "supply-chain", "name": "Safetensors", "sourceUrl": "https://github.com/huggingface/safetensors", "description": "Tensor storage format designed to avoid executable deserialization; it does not establish model provenance or behavioral safety", "directoryDescription": "Safetensors is included in the Awesome MLSecOps AI Supply-Chain Security directory. The community-maintained README describes it as: “Tensor storage format designed to avoid executable deserialization; it does not establish model provenance or behavioral safety.” Its MLSecOps relevance is the protection of model provenance, artifact integrity, dependencies, signing, bills of materials, registries, or delivery pipelines. The linked first-party source is the huggingface/safetensors repository on GitHub. A technical review should test the project's documented evidence across four criteria: Provenance and signing support, ML-BOM formats, Registry and CI integration, and Policy enforcement. Compare that evidence with the intended architecture and threat model. Catalog inclusion establishes relevance to this security category; it is not a certification, comparative ranking, or endorsement. Confirm current capabilities, maintenance, licensing, limitations, and deployment assumptions in the first-party documentation before adoption.", "source": "github.com/RiccardoBiosas/awesome-MLSecOps" } ]