Colorado legislators passed a law establishing guardrails on the use of artificial intelligence in the psychotherapy and mental heath space.
The new law, which went into effect in August, officially bans AI systems from providing psychotherapy services in place of a licensed, human practitioner. It also prohibits AI from generating therapeutic recommendations or treatment plans, requiring a practitioner to review and approve the output.
Calling AI tools “biased and unlicensed,” State Representative Javier Mabrey said in a statement that human care belongs at the center of medicine.
“Without this law, there is no clinical standard for AI chatbots, which can ultimately harm patients’ access to high-quality healthcare,” Mabrey said. “This law protects patients while still allowing providers to take advantage of AI tools for administrative tasks.”
Colorado Among First States To Restrict AI Use In Mental Health Care
Colorado is among the first states to pass comprehensive legislation addressing the use of consumer-facing AI in psychotherapy. Several families and organizations have sued major AI companies alleging that chatbots caused or contributed to mental health-related harms. State legislator Judy Amabile alluded broadly to the lawsuits in a public statement endorsing the bill.
“As policymakers, we cannot let chatbots, several of which are currently facing major lawsuits due to wrongful and horrifying deaths, replace certified mental health providers,” she said.
The new law also comes as more researchers and professional organizations push for greater oversight on AI in healthcare.
“AI therapy should not be used for youths and should not take the place of licensed clinical healthcare mental health professionals,” board-certified psychiatrist Dr. Judith Joseph tells HealthStack. “This is a recommendation by the American Psychological Association and the American Psychiatric Association.”
Joseph argues that AI tools should instead be leveraged to streamline research and administrative tasks. She adds the ideal balance is for the tools to decrease healthcare burdens to support human-centered care.
“It should be used to decrease healthcare professional burden and not to replace healthcare professionals,” Joseph says. “Establishing a relationship with a mental health professional who understands nuances in human behavior and child development and can implement evidence-based approaches is key to getting the best and safest patient outcomes.”
New Rules Explained
While the law limits AI from giving recommendations, it doesn’t prohibit it from supporting other aspects of mental health care. AI can assist providers with care, education, research or certain patient-directed tools.
When clinicians use AI to record or transcribe therapy sessions, they must inform patients in advance. Professionals must also explain how they will use AI and obtain each patient’s written, informed consent.
In addition, the new law restricts AI companies from marketing their tools for psychotherapy. Developers also cannot claim that patient data receives the same confidentiality protections as information shared with a therapist. They must also disclose that they are not a substitute for clinical care.
For Dr. Ramon Lizardo, a doctor who has held executive roles at health tech companies, the distinction is important.
“As someone who’s built and scaled health tech companies, I don’t read this as an anti-innovation law,” Lizardo tells HealthStack during an exclusive interview. “HB26-1195 doesn’t ban AI from mental health. It draws a line between AI replacing a clinician and AI supporting the broader care ecosystem.”
Developers Aim To Solve The Growing Mental Health Crisis
As consumer-facing mental health AI tools continue to deploy at record speed, developers argue they aim to address the growing crisis of Americans struggling to access timely mental health care.
Nearly 1 in 10 U.S. adults reported experiencing a mental health crisis within the past year. Patients in many parts of the country face long wait times, high costs, and limited access to mental health specialists. With AI tools, developers argue they can help fill gaps in care access.
Stanford researchers at the Human-Centered Artificial Intelligence Institute identified potential upsides to AI tools. They positioned them as supporting greater access and lower costs that could work alongside health systems.
Other industry executives have also promoted a hybrid approach for AI in healthcare. They argued that AI does not have to replace human care to make a meaningful impact on access.
Lizardo believes there’s a model where AI can serve as a force multiplier, helping human practitioners reach more people without replacing the human connection at the center of that care.
“I see AI becoming the connective tissue of a distributed care team rather than a stand-in for any single part of it,” he explains. “AI can match someone to the right peer mentor faster, flag escalating risk to a licensed clinician, cut the administrative burden that keeps clinicians from seeing more patients, and extend support into behavioral health deserts where there simply aren’t enough licensed providers. What it shouldn’t do is stand in for the clinical relationship itself. “
Regulation Remains A Moving Target
More than 140 state bills related to AI and mental health have been introduced, while federal legislation remains narrow. The growing number of state-level proposals could create an increasingly complex regulatory landscape for companies operating across multiple jurisdictions.
Lizardo predicts that Colorado’s approach could help shape how other states and the industry approach mental health AI.
“I’d bet against a 50-state patchwork of different products,” Lizardo says. “Companies build to the strictest applicable standard once, then deploy nationally. It’s cheaper, and it de-risks future regulation.”
Rather than fighting individual state requirements, Lizardo argues that companies could use Colorado’s standards as a baseline for how they develop and market mental health AI.
“The smart move for the industry isn’t to lobby this away state by state,” he explains. “It’s to treat ‘human-in-the-loop for clinical decisions, transparent disclosure, no deceptive equivalence claims’ as the design standard everywhere, and build the innovation on top of that foundation rather than around it.”
Whether that will happen, remains uncertain. One moving target is how regulators identify AI systems that cross the line into providing psychotherapy. Stanford Human-Centered Artificial Intelligence Institute noted that inconsistent definitions and evaluation standards can make it difficult to determine which AI tools qualify as mental health tools and how they should be regulated.
The uncertainty and rapidly changing regulatory landscape make this a closely watched area for clinicians and regulators.
