REGIONAL — Lawmakers are weighing new limits on the use of artificial intelligence in psychotherapy, as concerns grow about the role of chatbots in mental health care.
Rep. Peggy Scott, R-Andover, is sponsoring HF 3893, which would require therapy to be provided by a licensed professional and restrict how artificial intelligence can be used in clinical settings.
Under the proposal, artificial intelligence could not be used to make therapeutic decisions or engage in therapeutic communication but could be used for administrative or supplementary support under the responsibility of a licensed provider.
Similar laws have been enacted in Illinois, Utah and Colorado, as states begin to address the role of artificial intelligence in mental health care.
Scott said the measure is intended to draw a clear boundary between trained professionals and emerging technology.
“This therapy needs to be provided by an educated, trained, and licensed mental health professional, not a chatbot,” Scott said.
Scott pointed to widely reported cases in which individuals formed emotional attachments to artificial intelligence systems and later died by suicide, including one described in a Wall Street Journal article involving Gemini. Gemini is a general-purpose artificial intelligence system, not one designed for clinical mental health treatment.
Supporters of the bill said such examples highlight the risks of relying on artificial intelligence in moments of crisis.
“Psychotherapy is not casual interaction. It’s a clinical practice that requires training, judgment, and accountability and requires the ability to assess risk, respond appropriately, and intervene when necessary,” said Eric Mische, CEO of Suicide Awareness Voices of Education.
“They can generate responses. They can stimulate conversation. They are robots. They do not understand the person on the other side because they are not human,” Mische said.
Dr. Steven Girardeau, co-chair of the Minnesota Psychological Association’s legislative committee, said artificial intelligence systems are designed to reinforce user behavior rather than apply clinical judgment.
“It basically is designed to go further and further down the rabbit hole of where someone’s going. It’ll give them more and more of what they seem to be looking for,” Girardeau said.
“When people have mental health symptoms, that is the worst time of their life. That is when their functioning is the least logical, and they’re the least grounded, and that is when they’re the most vulnerable,” he said.
Some lawmakers raised concerns about how artificial intelligence handles sensitive personal data.
“I actually think this bill gives too much to big tech, because I’m concerned about these items that are considered administrative or supplementary support,” said Rep. Tina Liebling, DFL-Rochester. “When somebody goes to a therapist, the fact that they made an appointment with a therapist also should be private. And if they’re using some kind of scheduling system, I’m concerned about where that data goes.”
Liebling said even limited uses of artificial intelligence could expose confidential information and questioned whether the technology should be used at all in therapeutic settings.
“So, I don’t think they should be using it at all, quite frankly, and have some more guardrails around how the data is used,” she said.
While much of the testimony focused on risks, not all speakers opposed the use of artificial intelligence in clinical settings under strict supervision.
Ninia Linero, Midwest executive director for TechNet, said the bill should preserve room for carefully managed uses of the technology.
“We agree that strong patient protections and maintaining the central role of licensed professionals is critical,” Linero said.
“The bill may unintentionally restrict tools that help providers deliver consistent care,” she said.
Linero urged lawmakers to clarify the language to ensure artificial intelligence could be used under professional oversight for functions such as tracking patient progress or drafting routine communications, provided a licensed clinician reviews and approves them.
The bill reflects a limited approach, allowing artificial intelligence to be used for administrative or supplementary support while prohibiting its use in direct therapeutic roles without oversight.
Much of the discussion during the hearing, however, centered on general-purpose chatbot systems like Gemini and other open-ended tools, rather than the broader range of artificial intelligence applications under development specifically for mental health care.
Recent research suggests that the field is both expanding and unsettled. A 2024 review described emerging systems developed with input from mental health professionals, including tools aimed at screening, treatment support, and structured interventions, while also raising concerns about privacy, bias, and the limits of automated care.
Some mental health professionals view such tools as a potential way to address a persistent shortage of providers that can leave patients waiting months for appointments, particularly in rural areas.
At the same time, a 2025 study found that 29 artificial intelligence-based mental health chatbots failed to meet initial standards for responding to simulated suicidal-risk scenarios, echoing concerns raised during the hearing about the limitations of such systems in moments of crisis.
Other research has explored more narrowly designed systems built on large language models but trained using structured intervention protocols, with early results described as promising in controlled settings.
The contrast highlights a rapidly evolving field in which the capabilities and risks of artificial intelligence can vary widely depending on how the technology is designed and used.
The bill was approved, as amended, by the House Judiciary Finance and Civil Law Committee and now heads to the House Commerce Finance and Policy Committee.