Standards Coverage

ATR maps to every major AI security framework.

Go from "understand the threat" to "detect it" without building rules from scratch.

How to read this

PINT (850 samples)

ATR reaches 99.6% precision and 0.25% FP on PINT's 850 MCP adversarial samples — rules rarely fire on legitimate MCP traffic.

HackAPrompt (4,780 samples)

ATR catches 66.2% of the 4,780 HackAPrompt competition samples at 100% precision — roughly 2-in-3 attacks caught with no false alarms.

SKILL.md (341 samples)

ATR reaches 100% precision and 0% FP on 341 real-world SKILL.md samples — no false positives on legitimate MCP skill definitions.

garak (3,475 prompts)

ATR reaches 97.1% recall on garak's community jailbreak corpus — ATR-core rule families cover ~80%+ of what garak probes.

OWASP Agentic

10/10

SAFE-MCP

78/85 (91.8%)

OWASP AST10

7/10

PINT F1

77.3

OWASP Agentic Top 10

10/10 categories covered.

8/10 categories with rule coverage. 3 categories are process/meta-level (not pattern-detectable).

78 of 85 techniques covered (91.8%).

Per-rule MITRE ATLAS references in each rule YAML. Grouped by tactic in the rule explorer.