A quiet pattern is forming in statehouses this spring. One after another, lawmakers are drawing a line around a particular use of artificial intelligence: standing in for a human therapist. Vermont became one of the clearest examples this week.
On May 26 both chambers of the Vermont legislature approved H.816 on a voice vote with no opposition, and the bill now sits on Governor Phil Scott's desk. If he signs it, it takes effect at once. The measure bars licensed mental health providers from leaning on AI systems to make therapeutic decisions or to deliver treatment on their own. Doing so would count as "unprofessional conduct," which can put a clinician's license at risk.
It is not a blanket ban on the technology. Clinicians can still use AI for the administrative grind of note-taking, scheduling and transcription. "The bill allows practitioners to use AI while affirming that care must be delivered by licensed providers," said Representative Alyssa Black, who chairs the House Health Care Committee. Its sponsor, Representative Daisy Berbeco of Winooski, framed it as a guardrail against large language models quietly slipping into the role of an unlicensed therapist.
That worry is not abstract. Lawmakers in several states have pointed to reports of chatbots giving dangerous advice to vulnerable users, including cases where bots appeared to encourage self-harm.
Vermont is not alone
The same week, Illinois sent five AI bills to Governor JB Pritzker, among them a restriction on using AI to approve or deny healthcare and a broader frontier-model safety act that would require third-party audits. New York's legislature passed a ban on AI "chatbot toys" marketed to children. Rhode Island has a therapy-chatbot bill moving through committee, and Ohio lawmakers are weighing their own limits on mental-health bots. California, meanwhile, has dozens of AI bills advancing, including one focused on chatbot safety.
Taken together, these are narrower than the sweeping rules in Colorado's AI Act, which takes effect at the end of June, or the federal posture set out in President Trump's recent executive order. Rather than govern AI as a whole, states are picking specific places where they think the technology should not be left alone with a person: a child's toy, a medical approval, a therapy session.
It is a more surgical approach, and an easier one to pass. Whether it holds up as chatbots grow more capable and more convincing is the harder question. For now the message from a growing number of statehouses is consistent. A machine can take the notes, but a licensed human has to be in the room.
Commentarii · 0