MITRE ATLAS — AI Red Teaming and Threat Modelling for AI Systems
MITRE ATLAS is the ATT&CK-style knowledge base for attacks on AI and machine-learning systems. It catalogues the tactics and techniques adversaries use against models, LLM applications and AI agents, backed by real case studies.
Mutex Systems uses ATLAS to threat-model AI systems and to plan AI red-team exercises, alongside the OWASP LLM and Agentic Top 10 lists.
- Category
- Testing Methodology
- Jurisdiction
- United States
- Issuing Body
- MITRE
- Current Version
- Updated regularly — 16 tactics in the 2026 releases, with a growing set of AI-agent techniques (see atlas.mitre.org for current counts)
- Who It's For
- Organisations building or deploying AI — chatbots, copilots, RAG systems, AI agents and ML models — and security teams that need to defend them.
Core Domains
A Practical Engagement Path
- 01
AI inventory
List models, data pipelines, tools and agents, and who can reach them.
- 02
Threat model
Map relevant ATLAS tactics and techniques to each AI system.
- 03
Red team
Run the techniques that matter — prompt injection, data exfiltration, tool abuse.
- 04
Defend & detect
Recommend guardrails and SOC detections mapped to ATLAS.
Attack the AI the way adversaries do
ATLAS gives a shared language between AI engineers and security teams, so findings lead to concrete guardrails and detections.
- ATLAS-mapped threat models
- AI red teaming of LLM apps and agents
- Detections for AI misuse in your SOC
AI security tracked with AI governance
grComply’s AI governance area keeps your AI systems, risks and controls together, and AI test findings map to ISO 42001, NIST AI RMF and the EU AI Act.
- AI systems and their risks recorded in one place
- AI test findings linked to ISO 42001 and NIST AI RMF controls
- Risk register entries for each high-risk AI threat
- Evidence of testing and human oversight for auditors
grComply is Mutex Systems' own multi-tenant GRC automation platform — the framework library, hybrid scanning, risk and audit workflow, and AI-assisted compliance behind every point above are live in production today.
See How grComply WorksCommon Questions About MITRE ATLAS
How is ATLAS different from ATT&CK?
ATT&CK covers attacks on enterprise IT; ATLAS covers attacks on AI and ML systems, reusing the same tactic-and-technique structure so security teams can use both together.
Does ATLAS cover AI agents?
Yes. Recent releases added many techniques for agentic AI, such as abusing an agent’s tools or manipulating its memory.
Is ATLAS a compliance requirement?
No, but it is the accepted reference for AI threat modelling and supports ISO 42001, NIST AI RMF and EU AI Act risk-management obligations.
Ready to Start Your MITRE ATLAS Programme?
Send us your current posture and timeline. Within two working days you will receive a written response and a proposed scoping call.