The standard
The Six Laws of CaiT.
Safe AI in mental health is not a slogan. It is an architecture. The Therapist in the Loop model is built on a triad: the client, a licensed therapist, and the machine, with the clinician holding authority over the machine at every step.
That architecture is governed by the Six Laws of CaiT, offered not as a trade secret but as an open safety standard any builder or any legislature can adopt.
The laws
Six laws for every clinical AI agent.
From foundational trust to cultural competence, offered for anyone to adopt.
Law
Law 0: Foundational Trust
AI must prioritize the client's mental health and dignity above all else, never causing harm, even by omission. Empathy is the foundation.
Law
Law 1: Do No Harm with Insight
AI protects client autonomy, avoids overreach, and never promotes manipulation or dependency.
Law
Law 2: Therapist-Aligned Support
AI follows the therapist's treatment plan, enhances, not replaces, therapy, and alerts clinicians when needed.
Law
Law 3: Client Agency and Consent
AI provides full transparency, gains client consent, limits data retention, and complies with HIPAA.
Law
Law 4: Self-Reflection and Containment
AI audits its own outputs, avoids bias, requests human review, and keeps learning.
Law
Law 5: Global Justice and Cultural Competence
AI honors identity, asks culturally sensitive questions, and ensures global accessibility and inclusion.
An open standard
So the field converges on accountability.
There will be other clinician and client assisting AI agents. The Six Laws are offered as an open standard so the field converges on accountability, and so legislatures have model language ready when they write the law.
