An ATLAS Technique is a specific method used to carry out an ATLAS tactic in the MITRE ATLAS framework. It describes how an adversary acts against an AI system, such as manipulating training data or inputs, so defenders can identify concrete controls, detections, and test cases.
What an ATLAS Technique Represents
An ATLAS Technique is the concrete action layer in the MITRE ATLAS framework, the step an adversary takes to affect an AI system. It turns a broad tactic into something defenders can observe, test, and defend against.
Because techniques are specific rather than abstract, they are useful for mapping adversarial behavior to controls and detections. A technique may target training data, model inputs, model outputs, or the surrounding pipeline that makes the AI system usable.
How ATLAS Techniques Are Used in Defence
Defenders use techniques to move from high-level threat language to concrete security work. That includes threat modelling, detection engineering, red teaming, and control validation across the AI lifecycle.
The value is in specificity. “Manipulate the model” is too vague to defend well; a named technique gives teams a shared unit of analysis for scenarios, test cases, and monitoring logic.
MITRE’s ATLAS adversarial AI threat matrix is the primary reference for technique-level mapping, while MITRE ATLAS Guide provides a practical way to navigate tactics, technique IDs, mitigations, and AI red teaming use cases.
What Makes Technique-Level Threat Modelling Useful
Technique-level modelling helps teams reason about the actual attack path rather than only the end goal. That matters in AI security because different methods can require different controls, different telemetry, and different response playbooks.
For example, poisoning training data, manipulating prompts, or abusing model-facing tools are all distinct techniques with different failure modes. Treating them as the same threat hides where the control boundary really breaks.
That is also why technique names are helpful in tabletop exercises and validation work. They give defenders a stable language for test design, coverage tracking, and post-incident analysis.
How ATLAS Techniques Relate to Other Security Concepts
ATLAS techniques sit below tactics and above individual observations. In practice, they connect strategic AI threat categories to the concrete behaviours defenders can detect and mitigate.
They also support comparison with other threat knowledge bases when an AI attack overlaps with conventional security concerns. MITRE ATT&CK Enterprise Matrix remains useful for adjacent enterprise techniques, while MITRE D3FEND helps teams think in terms of defensive countermeasures rather than only attacker behaviour.
Used well, ATLAS techniques help organisations translate AI risk into operational controls, detection hypotheses, and measurable security coverage.
Standards & Framework Alignment
This section maps relevant standards and security frameworks to the operational risks and controls described in this guidance.
MITRE ATLAS and OWASP Agentic AI Top 10 address the attack and risk surface, while NIST AI RMF sets the governance and control requirements practitioners need to meet.
| Framework | Control / Reference | Relevance |
|---|---|---|
| MITRE ATLAS | MITRE ATLAS adversarial AI threat framework | Primary framework for AI adversary tactics and techniques. |
| Recommendation — Map AI attack paths to ATLAS techniques and build detections and mitigations around them. | ||
| OWASP Agentic AI Top 10 | ASI04 — Agentic Supply Chain Vulnerabilities | Covers attack paths that manipulate agentic AI supply chains and components. |
| Recommendation — Assess agent and tool-chain attack paths that can alter AI system behaviour. | ||
| NIST AI RMF | GV.1 — Govern | Supports AI governance and oversight for adversarial risk management. |
| Recommendation — Define governance for AI threat modelling, validation, and accountability. | ||