Cybernion
← Insights

Security

What Is MITRE ATLAS?

30 Aug 2026 · Updated 30 Aug 2026

MITRE ATLAS (Adversarial Threat Landscape for Artificial-Intelligence Systems) is a public knowledge base of real-world tactics and techniques adversaries use to attack AI and machine learning systems. It is the AI-focused counterpart to MITRE ATT&CK, using the same structured tactics-and-techniques format. It covers evasion, data poisoning, model extraction, inference attacks and generative-AI threats like prompt injection, and organisations use it to guide AI threat modelling, red teaming and risk assessment.

As organisations rush to deploy AI, a natural question follows: how exactly are these things attacked? MITRE ATLAS is the structured answer, a catalogue of how AI systems are attacked in the real world.

What is MITRE ATLAS and how does it relate to ATT&CK?

MITRE ATLAS, Adversarial Threat Landscape for Artificial-Intelligence Systems, is a public knowledge base that documents the tactics and techniques adversaries use to attack AI and machine learning systems, drawn from real-world attacks and security research. It gives defenders a structured, shared way to understand and describe how AI systems are attacked, rather than treating AI security as a vague new frontier.

ATLAS is deliberately the AI-focused counterpart to the widely used MITRE ATT&CK framework. ATT&CK catalogues adversary tactics (the attacker’s goals, such as reconnaissance, initial access, or impact) and techniques (how they achieve them) against IT systems in general. ATLAS applies the same tactics-and-techniques structure specifically to attacks on AI and machine learning. Because it mirrors ATT&CK’s familiar format, security teams that already use ATT&CK can adopt ATLAS readily, and the two are complementary: an attack on an AI-powered system may combine conventional ATT&CK techniques to reach the system with ATLAS techniques to attack the model itself.

What it covers and how organisations use it

ATLAS covers the distinctly AI-specific attack techniques. These include evasion (crafting inputs that fool a model into a wrong output), data poisoning (corrupting the training data so the model learns the wrong thing), model extraction or theft (reconstructing a proprietary model by querying it), membership inference (working out whether particular data was in the training set), and, increasingly, attacks on generative AI and large language models such as prompt injection and jailbreaks. These are organised into tactics representing the attacker’s objectives across the lifecycle of an attack on an AI system.

Organisations use ATLAS in several practical ways: to understand the threats their AI systems actually face, to guide AI red teaming and testing by giving testers a catalogue of techniques to attempt, to inform risk assessments and threat modelling for AI systems, and as a common vocabulary so teams can describe AI attacks precisely. It complements the OWASP LLM Top 10, which focuses on the most critical risks in LLM applications, by providing a broader, attacker-centric map of techniques. Together they help teams move from a general unease that AI can be attacked to a structured understanding of specific techniques and their mitigations, which is the foundation of building and governing AI securely. Where an organisation is securing or governing AI systems, we can help through ISO 42001 advisory and penetration testing.

Frequently asked questions

What is MITRE ATLAS?

MITRE ATLAS (Adversarial Threat Landscape for Artificial-Intelligence Systems) is a public knowledge base that documents the tactics and techniques adversaries use to attack AI and machine learning systems, based on real-world attacks and research. It is modelled on the well-known MITRE ATT&CK framework but focused on AI-specific threats, giving defenders a structured, shared way to understand and describe how AI systems are attacked.

How does ATLAS relate to MITRE ATT&CK?

ATLAS is the AI-focused counterpart to ATT&CK. ATT&CK catalogues adversary tactics and techniques against IT systems generally, while ATLAS applies the same structured tactics-and-techniques approach specifically to attacks on AI and machine learning, such as evading models, poisoning training data, or extracting a model. They are complementary, and ATLAS deliberately mirrors ATT&CK’s format so security teams already familiar with ATT&CK can adopt it easily.

What kinds of AI attacks does ATLAS cover?

ATLAS covers AI-specific attack techniques including evasion (crafting inputs that fool a model), data poisoning (corrupting training data), model extraction or theft, membership inference (learning whether data was in the training set), and increasingly attacks on generative AI and large language models such as prompt injection. It organises these into tactics representing the attacker’s goals, from reconnaissance through to impact.

How do organisations use MITRE ATLAS?

Organisations use ATLAS to understand the threats their AI systems face, to guide AI red teaming and testing, to inform risk assessments and threat modelling for AI, and as a shared vocabulary for describing AI attacks. It helps teams move from a vague sense that AI can be attacked to a structured understanding of specific techniques and mitigations, which supports building and governing AI systems more securely.


Written by Gaurav Vikash, an ASD endorsed IRAP assessor and senior cyber security leader with 18 years of experience across Australia, the UK and Asia, including CISO and senior security leadership roles. He holds CISSP, CISA, CISM and CRISC and is an ISO 27001 Lead Auditor and ISO 42001 Lead Implementer, and speaks regularly at industry conferences.

Talk to us.

Sources:

  1. MITRE ATLAS, MITRE, 2026
  2. MITRE ATT&CK, MITRE, 2026

Last updated: 30 August, 2026

Talk to us about your engagement

A scoped proposal within one business day.