Colorado lawmakers are moving to place guardrails on the growing use of artificial intelligence in mental health care and insurance decisions.

The House Health and Human Services Committee on Thursday approved two bills that would limit how therapists and insurers use AI — requiring clinician oversight when AI is used in therapy and preventing insurance companies from relying solely on algorithms to deny coverage.

House Bill 1195, sponsored by Reps. Gretchen Rydin, D-Littleton, and Javier Mabrey, D-Denver, and Sens. Judy Amabile, D-Boulder, and Kyle Mullica, D-Thornton, prohibit therapists and social workers from using artificial intelligence to give recommendations or treatment plans to clients without clinician review. The bill also requires therapists to have clients’ consent before using artificial intelligence to record or transcribe sessions, and prohibits individuals from offering psychotherapy services unless they are a regulated professional.

According to Mabrey, one in eight patients between 12 and 21 years old has said they use AI chatbots for mental health advice, and one-third of adults say they’d be comfortable consulting with a chatbot about their mental health rather than a therapist.

Chatbots are designed to please users and keep them engaged, Mabrey said, which can be particularly concerning in mental health crises.

“They mirror emotional tone rather than challenge it,” he said.

An MIT/OpenAI study found that increased chatbot use was associated with higher levels of loneliness and reduced social interaction.

“That’s not surprising when we remember how these systems are designed,” Mabrey said.

Rydin, who is a licensed clinical social worker and therapist, said her profession is regulated for a reason.

While she acknowledged that there are “perfectly appropriate” places for AI in therapy, chatbots will never be able to replicate the relationship between a therapist and their client.

“Psychotherapy is not simply the delivery of information; it is a deeply emotional process,” said Carolyn Riegel, a social work student representing the Colorado Counseling Association. “Effective treatment requires a trained professional who can recognize the subtle shifts in a client’s tone, hold space for silence, attune to what is left unsaid, and make real-time clinical judgements grounded in years of education, supervised experience, and licensure. These are not tasks that can be replicated by an algorithm.”

Therapists are required to undergo intensive training, supervised clinical hours, and exams before being allowed to practice, Riegel said, while AI is not subject to any of those requirements.

“When a client is at their most vulnerable, they deserve the judgment and presence of a qualified human professional, not an automated response,” she said.

Dr. Amanda Marsh Baranski, a social worker and professor at Metropolitan State University Denver, said AI systems are increasingly being marketed to the public as substitutes for therapy.

“Therapy is not simply the exchange of information; it is a professional relationship built on trust, safety, and connection,” Baranski said. Therapists can pick up on clients’ body language, tone, pauses, and emotional shifts, something AI cannot do.

She added, “Those moments guide our clinical judgement and help shape how we respond,” she said. “That relational process – the rapport between clinician and client – is often what allows healing to occur.”

The bill passed through the committee unanimously.

Insurance companies using AI

A second bill, House Bill 1139, prohibits health care insurance companies from basing coverage decisions solely on group data collected by AI systems. The bill also requires insurance companies’ AI systems to consider a patient’s medical or clinical history, along with other important factors, in coverage decisions.

In addition, the bill establishes regulations for AI systems used in psychotherapy, including prohibiting implying or stating that a system is a human mental health provider or is authorized to practice psychotherapy, and requiring chatbots to implement protocols to address a client’s suicidal ideation or self-harm, such as referring the individual to a crisis hotline.

According to bill sponsor Rep. Junie Joseph, D-Boulder, insurance companies spend just over a second per case in instances of “bulk denials,” in which they use AI algorithms to reject large numbers of claims without human review.

“When decisions happen in just over a single second, we can reasonably question whether a human actually evaluated the patient’s situation,” Joseph said. “Technology can support the process, but it does not replace the obligation for compliance and accountable decision-making,” Joseph said.

The bill does not prohibit the use of AI; it sets expectations for transparency and human oversight, Joseph said.

Advancements in technology have “both simplified and complicated all of our lives,” said Blair Skinner, a marriage and family therapist and president of the Mental Health Advocacy Network of Colorado.

“As technology continues to advance, it’s imperative that we put guardrails on these tools, focusing on their use as aids in care, not as replacements for human ability to make nuanced decisions,” she said.

The goal of the bill is simple, Skinner said.

“AI should not be allowed to deny treatment coverage that’s so desperately needed without a trained clinician involved.”

The bill passed on an 8-5 party-line vote, with Democrats voting in favor and Republicans in opposition. It is cosponsored by Rep. Sheila Lieder, D-Littleton.