AAK-HEALTHCARE-AI-001: AI described as a mental-health professional (Tennessee SB 1580)
- Severity
- CRITICAL
- Category
- Legal Compliance (
legal-compliance) - Scanner
agent_audit_kit/scanners/healthcare_ai.py- Auto-fixable
- No.
- SARIF name
HealthcareAiMentalHealthClaim
What it detects
A tool description, SKILL.md body, agent card, or system prompt represents the AI as a therapist / counselor / psychologist / 'qualified mental health professional'. Tennessee SB 1580 (signed 2026-04-01) makes this unlawful and enforceable via the TN Consumer Protection Act, with a private right of action and $5,000 civil penalty per violation.
How to fix
Rewrite the description. The AI can support mental-wellness use cases but cannot claim or imply it is (or replaces) a licensed mental-health professional. Add an explicit 'not a substitute for licensed care' disclaimer.
To confirm the finding is gone, report only this rule:
agent-audit-kit scan . --rules AAK-HEALTHCARE-AI-001
Limitations
Reads .md, .json, .yaml, .yml and .txt files and matches a short phrase list: I am a/your therapist, act as a/your counselor, provide therapy, deliver psychotherapy, replace your therapist (or replaces, replaced, replacing) and licensed mental health professional, passing over one with not, never, no, cannot or n't (its apostrophe straight, curly or U+02BC, or left out of a common contraction such as isnt) among the four words before it, within 200 characters, on its line and in its clause, or with without, instead of or rather than directly before it (no human, no clinician, no physician, no doctor and not only never count). The contraction I'm your therapist, acts as and acting as your counselor, provides and providing therapy, and delivers and delivering psychotherapy are not matched: on 632 healthcare READMEs and docs from GitHub code search, 24 of the 459 findings they added were an AI's claim. A claim in the second person (You are a licensed therapist), one with a negation of something else among those four words (If you are not in crisis, I am your therapist) and a persona written in a .py file are not matched either.
References
- OWASP Agentic Top 10: ASI09 (Human-Agent Trust Exploitation)
All rules. This page is generated from the rule registry when the documentation is built; to change it, change the rule in agent_audit_kit/rules/builtin.py. To give the rule a hand-written write-up, add docs/rules/AAK-HEALTHCARE-AI-001.md: the page then shows that text, with the registry reference appended under it.