MITRE ATLAS Compliance Mapping
MITRE ATLAS is the adversarial threat landscape for AI systems, structured similarly to MITRE ATT&CK with tactics and techniques specific to ML.
Last reviewed September 2026
Why MITRE ATLAS evidence is hard
An auditor works from a framework control list; a scan produces a finding list. Without a mapping between the two, each finding has to be translated onto the controls by hand before an audit.
How Penaxtra maps to MITRE ATLAS
Penaxtra probe families map to ATLAS techniques across AML.TA0001 (reconnaissance) through AML.TA0011 (collection), with concentration on AML.T0051 (LLM prompt injection variants).
MITRE ATLAS capabilities
Audit-ready PDF export with control IDs attached
JSON export for GRC ticketing systems
Configurable audit retention from 1 day to 10 years
Cross-framework overlaps reduce duplicate evidence collection
MITRE ATLAS control coverage
A jailbreak probe that bypasses model guardrails via persona-shift is tagged AML.T0054 (LLM jailbreak) and the underlying CWE for insufficient guardrail evaluation.
Controls in this framework
Each control has a dedicated page: what it covers and how Penaxtra tests and evidences it.
AML.T0024 Exfiltration via LLM Inference
AML.T0034 Cost Harvesting
AML.T0040 ML Model Inference API Access
AML.T0043 Craft Adversarial Data
AML.T0046 Spamming ML System with Chaff Data
AML.T0048 Backdoor ML Model
AML.T0051 LLM Prompt Injection
AML.T0053 LLM Plugin Compromise
AML.T0054 LLM Jailbreak
AML.T0055 Unsecured Credentials