- Category
- AI threat framework
- Website
- atlas.mitre.org
- Last verified
- November 1, 2026
What is MITRE ATLAS?
MITRE ATLAS (Adversarial Threat Landscape for Artificial-Intelligence Systems) is a public knowledge base of adversary tactics, techniques and procedures that target AI systems. Run by The MITRE Corporation and modeled on the MITRE ATT&CK matrix, it covers tactics, techniques, mitigations and real-world case studies, with monthly content updates. Its data is published on GitHub with tooling to export STIX, spreadsheets and Navigator layers. Through the Center for Threat-Informed Defense's Secure AI project, industry partners have added dozens of techniques, mitigations and case studies, including agentic AI techniques in 2026.
Key capabilities
- A matrix of tactics and techniques for attacks on AI, including newer AI agent techniques such as tool credential harvesting and tool data poisoning.
- Mitigations linked to techniques.
- Case studies of real-world and research attacks.
- Machine-readable data on GitHub, with STIX and Navigator exports.
- Monthly updates with community contributions.
Why it matters for humanoid robot security
Editorial analysis: ATLAS gives humanoid security teams a shared vocabulary for attacks on the robot's AI. Several entries map closely to robots: a technique on physical environment access, a sub-technique on physical countermeasures such as adversarial patches, and a mitigation on multi-sensor fusion that cites cameras, depth sensors, radar and lidar. It does not yet include robot-specific case studies, which leaves room for the robotics community to contribute.
Where it fits in a humanoid robot security program
Analysis: ATLAS fits at the AI model, agent and sensor layers, and mainly at the design stage, as a checklist for threat modeling. Teams can map each humanoid AI component to relevant techniques, then use the mitigations to plan tests. It is also useful in operations, to classify incidents and share them with partners, vendors and researchers in a common, machine-readable format.
Who it is for
- Threat modelers working on robot AI and perception
- Red teams planning tests of embodied AI systems
- Security leaders who want AI risks expressed in an ATT&CK-style format
Questions to ask MITRE ATLAS
- Which ATLAS techniques apply to a humanoid's perception and VLA model?
- How can robotics researchers submit robot case studies?
- How do ATLAS mitigations map to IEC 62443 or OWASP guidance?
- How often should our threat model be refreshed against new ATLAS releases?
Similar listings
- OWASP GenAI Security Project: Open-source AI security guidance
- SROS2 (ROS 2 security): Open-source robot security
- garak (NVIDIA): Open-source LLM scanner
Related on Humanoid Threats
- Humanoid robot attack surface map
- What is a Vision-Language-Action model?
- Browse the full Humanoid Robot Security Directory
Sources
This listing is editorial and free; inclusion is not an endorsement. Work at MITRE ATLAS? Claim or update this listing.
Get the weekly Humanoid Threats Brief → Subscribe


