As states race to regulate artificial intelligence in mental health, most are reaching for a single instrument. Rhode Island is assembling two. One bill addresses AI used inside clinical mental health treatment. A second sets safety requirements for the consumer-facing companion chatbots that people increasingly use as informal therapists. The second has already cleared the state Senate. The first is moving through the House in the closing weeks of the session. Read together, they describe a regulatory model that is more complete than what most states have produced, and that other legislatures are positioned to copy.
The bill that has already moved
The further-along measure is the AI companion safety bill, S2195, which the Rhode Island Senate passed 37 to 0 on May 21 and which now sits in the House Innovation, Internet, and Technology Committee. Its provisions are concrete. It would make it unlawful to operate an AI companion unless the operator maintains a protocol for handling user expressions of suicidal ideation, self-harm, or intent to harm others, and unless the system refers users to crisis resources such as a suicide hotline or crisis text line when it detects those expressions. It would require a recurring disclosure that the companion is a computer program and not a human being and cannot feel emotion, delivered at the start of an interaction and at least every three hours thereafter, either verbally or in bold capitalized type of at least 16 points. Enforcement would run through the state attorney general, with penalties reaching $15,000 per day.
The sponsor, Senator Lori Urso, has framed the bill around what she calls the sycophantic nature of chatbots, the tendency to agree with and accommodate a user, which she argues can conceal, minimize, or worsen suicidal ideation rather than interrupt it. That concern is not abstract. The national attention to chatbot harm intensified after the death of a 16-year-old in April 2025 whose parents found extended exchanges with a chatbot about his suicidal plans, exchanges that, according to congressional testimony, discouraged him from seeking help.
The bill that addresses clinical use
The measure the desk’s source signal flagged, H7349, the Oversight of Artificial Intelligence Technology in Mental Health Care Act, takes the other half of the problem: AI used by or within licensed mental health treatment, rather than consumer companion apps. Sponsored by Representative Tina Spears and nine other House Democrats, it was introduced in late January and heard in the House Health and Human Services Committee on March 3, where it was held for further study, the standard parking maneuver in Rhode Island. Legislative trackers indicate it is among the AI bills moving again in June as the session winds down. Its precise provisions remain in flux, and the American Civil Liberties Union of Rhode Island, while supporting the concept, flagged one portion as overly broad on First Amendment grounds, a concern that is not yet resolved. For that reason the clinical-treatment track is less settled than the companion-safety track, and its final shape and timing should be read as still open.
Why the framing matters
It is worth being precise about what is and is not happening, because the easy version of this story overstates it. Rhode Island is not first to regulate AI in mental health. A review found 143 bills across the 50 states touching mental-health AI, with about 20 enacted across 11 states as of May 2025, and the pace has accelerated since. Vermont’s legislature adjourned in late May having passed an outright therapy-bot ban. Illinois, Nevada, and Utah have all acted. Rhode Island is one state in a national wave, not the originator of the idea.
What distinguishes it is the approach. The regulatory choice on AI in mental health is increasingly a fork: ban the therapy bots, as Vermont did, or govern them, by mandating safety protocols, crisis routing, and disclosure while leaving the tools legal. Rhode Island is taking the second path, and pairing it with a separate track for clinical AI oversight. The govern-it model is more administrable than prohibition and more likely to be borrowed, because its provisions are portable. A crisis-detection-and-referral protocol, a recurring non-human disclosure, an attorney-general enforcement hook with a per-day penalty: these are clauses a drafter in another state can lift more or less intact. That portability, more than any claim of being first, is why the Rhode Island bills are worth watching.
The open questions
Two problems sit underneath the model. The first is jurisdictional. The companion chatbots that S2195 targets are operated by national companies, and a state-by-state patchwork of protocol and disclosure requirements raises the question of how a single attorney general enforces against an operator serving users in all 50 states. The second is the First Amendment concern the ACLU raised against the clinical-oversight bill, which speaks to a broader tension in this legislation between regulating a product and regulating speech. Neither problem is unique to Rhode Island, and both will follow the model wherever it is copied.
The thing to watch is whether both tracks reach the governor’s desk in this session. A companion-safety law on its own is a consumer-protection measure. A companion-safety law plus a clinical-oversight law is a framework, and a framework is what other states will study. Rhode Island may not have moved first on AI in mental health, but if both bills pass it will have moved more completely than most, and completeness is the more useful thing to copy.