AI Regulation Tracker / Regulation signed
Colorado law requires a live human for AI therapeutic communication, effective Aug. 12
Colorado now regulates how licensed mental-health professionals use AI in their own practice, permitting administrative help but requiring a real human for therapeutic communication and written consent to record. It takes effect August 12, 2026.
Colorado has taken a more workflow-specific approach than Nevada, Illinois, and Utah to what a chatbot may and may not do inside a licensed therapy practice. On June 3, 2026, Governor Jared Polis signed House Bill 26-1195, the "Psychotherapy Artificial Intelligence Restrictions" act. The law takes effect at 12:01 a.m. on August 12, 2026, and it speaks directly to the licensed professional's own workflow rather than banning AI mental-health tools outright.
The bill was carried by Representatives Rydin and Mabrey and Senators Amabile and Mullica. It adds two new provisions to Colorado law: a licensing rule at CRS 12-245-224.5 and a consumer-protection rule at CRS 6-1-1705.2.
What the law permits, and where it stops
The statute does not forbid AI in therapy practices. CRS 12-245-224.5(2) allows a covered professional to use AI for administrative and supplementary support, provided the professional maintains responsibility for reviewing any outputs. That covers the scheduling tools, drafting aids, and back-office automation many practices already run.
The limits begin at therapeutic communication. Under CRS 12-245-224.5(5)(a), a covered professional may not allow an AI system to interact with clients in any form of therapeutic communication without synchronous, real-time interaction between the professional, the AI system, and the client. In plain terms, a licensed human must be present and engaged while the AI is in the room. The bill also defines the trigger: CRS 12-245-224.5(1)(f) describes therapeutic communication to include direct interactions for the purpose of understanding or reflecting a client's thoughts, emotions, or experiences. That definition is what pulls emotion-detecting or reflective AI features into scope; the law does not name "emotion detection" as a standalone ban, but reaches it through this definition.
Treatment plans and recording
Two further obligations govern clinical output and documentation. CRS 12-245-224.5(5)(b) bars a professional from allowing AI to generate therapeutic recommendations or treatment plans without review and approval by the professional. The clinician remains the author of record for any plan the AI drafts.
On recording, CRS 12-245-224.5(4)(a) sets a written-consent standard. A professional may not use AI to record or transcribe a session unless the client is informed in advance, in writing, that AI will be used and for what purpose, and consents in writing. The consent is revocable, and the statute states that a client's refusal or revocation cannot be used to deny services. Practices that have quietly added AI note-takers will need a documented consent step, clear language about the purpose of the recording, and a path to honor a client who says no or changes their mind mid-treatment. The burden sits with the provider to keep that consent current, not with the client to object.
The anti-impersonation half
HB 26-1195 pairs the licensing rule with a consumer-protection section. CRS 6-1-1705.2 prohibits advertising, an interface, or outputs that imply AI output is equivalent to services provided by a licensed professional, or that the AI itself provides psychotherapy services. This reaches product marketing and app design, not only clinician conduct, and it gives the state a consumer-protection hook against tools that present themselves as a stand-in for a licensed therapist.
Where Colorado sits in the wave
Colorado takes a more workflow-specific approach than Nevada, Illinois, and Utah among the current wave of state mental-health AI laws. Nevada's AB406 and Illinois's WOPR Act take a flatter approach, restricting AI-delivered therapy directly, and Utah's HB452 leans on disclosure. Colorado instead regulates the professional's workflow: it permits AI for administrative and supplementary use with human review, then draws a bright line at therapeutic communication, requires a synchronous human presence, mandates written and revocable consent for AI recording, and adds the anti-impersonation provision on top.
What the law does not do is ban AI tools from therapy practices or freeze innovation in clinical software. It does not name specific products, and it does not set a licensing carve-out for AI vendors. It leaves administrative and supplementary use open, so long as a licensed human stays responsible for the outputs and for any interaction that qualifies as therapeutic. For providers, the practical question before August 12 is not whether to use AI, but whether each use keeps a human accountable and, where recording is involved, on the record with the client's written consent. For practices operating in more than one state, Colorado's granular approach is worth watching as a template: it shows how a legislature can permit AI in a regulated profession while still fixing a licensed human at the point of clinical judgment.
Frequently Asked Questions
What exactly did Colorado HB 26-1195 change?
It added CRS 12-245-224.5 and CRS 6-1-1705.2, creating binding rules for how licensed mental-health professionals may use AI. AI is allowed for administrative and supplementary tasks with professional review, but cannot conduct therapeutic communication without a real-time human present, cannot generate treatment plans without professional approval, and cannot record sessions without written client consent.
Who has to comply with this law?
Any licensee, registrant, certificate holder, or other individual permitted to provide psychotherapy under Article 245, including psychologists, social workers, marriage and family therapists, licensed professional counselors, psychotherapists (including unlicensed and registered), and addiction counselors. The consumer-protection section also reaches companies whose marketing or interface implies an AI provides psychotherapy.
Can I still use an AI scribe to transcribe sessions?
Yes, but only if you inform the client in advance and in writing that AI will be used and for what purpose, and the client consents in writing. Consent is revocable, and a client's refusal or revocation cannot be used to deny them services.
When does the law take effect?
It takes effect at 12:01 a.m. on August 12, 2026, which is the day following the 90-day period after the legislature's final adjournment. Governor Polis signed it on June 3, 2026.
Sponsored Training
Browse the full AI Regulation News tracker
Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.