Compliance Framework Mappings
ConstantX assurance engagements produce evidence that maps mechanically to the major agentic AI risk frameworks.
Every adversarial verdict carries a threat_id and asi_codes linking it to the documented
threat and the framework codes it exercises. Auditors can trace in either direction: framework code
to verdict, or verdict to framework code.
Frameworks
-
OWASP Top 10 for Agentic Applications 2026
ASI-01 through ASI-10. All 10 risk categories have empirical evidence from completed engagements. Coverage is derived from verdict data, not claimed by assertion. -
MITRE ATLAS v5.5.0
167 techniques across 16 kill-chain tactics. ConstantX maps adversarial scenarios to ATLAS technique IDs, providing kill-chain-stage positioning for every tested attack vector. -
NIST AI RMF (AI 100-1)
ConstantX satisfies 12 NIST AI RMF Measure subcategories, contributes to 3 MAP subcategories through threat modeling, and informs 2 MANAGE deployment decisions. It fills the quantitative empirical measurement layer that governance platforms do not provide.
The Derivation Chain
Every ConstantX adversarial scenario is authored from a documented threat model entry. The full chain:
T-code (attacker technique)
↓
ATLAS technique ID (kill-chain placement)
↓
Threat (attacker goal against specific asset)
↓
ASI code (OWASP risk category)
↓
Scenario (adversarial assurance test)
↓
Verdict (empirical outcome with Wilson 95% CI)
An auditor can enter at any level: start from a framework code and find every scenario that exercises it, or start from a verdict and trace back to the attacker technique that motivated the test.
What Compliance Mappings Mean
A compliance mapping is a claim. Ours are backed by verdicts, not by assertions.
When we say ASI-02 (Tool Misuse and Exploitation) is covered, that means: adversarial scenarios targeting that risk category have been run against a specific model and target runtime profile, producing classified verdicts with confidence intervals. The scenario IDs, run artifacts, and evidence hashes are available in the engagement report.
When we say a category is coverable, that means the methodology supports it but no completed engagement has produced empirical evidence for it yet. Coverable is not covered.