ATLAS Matrix

The MITRE ATLAS matrix (v5.6.0) — 16 tactics read left to right as the adversary lifecycle. Each cell is a technique, shaded by how many curated attacks in this catalog exercise it. Click any technique to see the mapped attacks and its ATLAS reference.

Only techniques with curated attacks

74 / 111 techniques covered 0 1 2 3+

Reconnaissance

AML.TA0002 · 2 attacks

Search Open Technical Databases Search Open AI Vulnerability Analysis Search Victim-Owned Websites Search Application Repositories Active Scanning2 Gather RAG-Indexed Targets Gather Victim Identity Information Search Open Websites/Domains

Resource Development

AML.TA0003 · 13 attacks

Acquire Public AI Artifacts1 Acquire Infrastructure2 Obtain Capabilities1 Develop Capabilities Publish Poisoned Datasets1 Poison Training Data4 Establish Accounts Publish Poisoned Models4 Publish Hallucinated Entities1 LLM Prompt Crafting1 Retrieval Content Crafting1 Stage Capabilities Publish Poisoned AI Agent Tool1

Initial Access

AML.TA0004 · 20 attacks

AI Supply Chain Compromise9 Valid Accounts Evade AI Model4 Exploit Public-Facing Application2 Phishing1 Drive-by Compromise Prompt Infiltration via Public-Facing Application5

AI Model Access

AML.TA0000 · 5 attacks

AI Model Inference API Access3 Physical Environment Access2 Full AI Model Access AI-Enabled Product or Service

Execution

AML.TA0005 · 23 attacks

User Execution7 Command and Scripting Interpreter1 LLM Prompt Injection17 AI Agent Tool Invocation3 AI Agent Clickbait Deploy AI Agent

Persistence

AML.TA0006 · 15 attacks

Manipulate AI Model2 Poison Training Data4 LLM Prompt Self-Replication RAG Poisoning2 AI Agent Context Poisoning2 Modify AI Agent Configuration2 Prompt Infiltration via Public-Facing Application5 AI Agent Tool Data Poisoning AI Agent Tool Poisoning

Privilege Escalation

AML.TA0012 · 6 attacks

Valid Accounts AI Agent Tool Invocation3 LLM Jailbreak3 Escape to Host1

Defense Evasion

AML.TA0007 · 13 attacks

Evade AI Model4 LLM Jailbreak3 LLM Trusted Output Components Manipulation2 LLM Prompt Obfuscation2 False RAG Entry Injection Impersonation1 Masquerading Corrupt AI Model Modify AI Agent Configuration2 Manipulate User LLM Chat History1 Delay Execution of LLM Instructions Virtualization/Sandbox Evasion Exploitation for Defense Evasion AI Supply Chain Rug Pull1 AI Supply Chain Reputation Inflation

Credential Access

AML.TA0013 · 7 attacks

Unsecured Credentials5 RAG Credential Harvesting1 Credentials from AI Agent Configuration OS Credential Dumping AI Agent Tool Credential Harvesting1 Exploitation for Credential Access2

Discovery

AML.TA0008 · 6 attacks

Discover AI Artifacts1 Discover AI Model Ontology1 Discover AI Model Family Discover LLM Hallucinations1 Discover AI Model Outputs Discover LLM System Information1 Cloud Service Discovery Discover AI Agent Configuration1 Process Discovery1

Lateral Movement

AML.TA0015 · 2 attacks

Phishing1 Use Alternate Authentication Material1

Collection

AML.TA0009 · 4 attacks

AI Artifact Collection1 Data from Information Repositories Data from Local System1 Data from AI Services2

AI Attack Staging

AML.TA0001 · 7 attacks

Create Proxy AI Model1 Manipulate AI Model2 Verify Attack Craft Adversarial Data4 Generate Deepfakes1 Generate Malicious Commands

Command and Control

AML.TA0014 · 3 attacks

Reverse Shell1 AI Service API AI Agent2

Exfiltration

AML.TA0010 · 14 attacks

Exfiltration via AI Inference API3 Exfiltration via Cyber Means2 Extract LLM System Prompt1 LLM Data Leakage1 LLM Response Rendering1 Exfiltration via AI Agent Tool Invocation8

Impact

AML.TA0011 · 18 attacks

Evade AI Model4 Denial of AI Service2 Erode AI Model Integrity1 Cost Harvesting1 Spamming AI System with Chaff Data1 External Harms11 Erode Dataset Integrity2 Data Destruction via AI Agent Tool Invocation1 Machine Compromise