The Government Accountability Office said in a report released Nov. 20 that the Food and Drug Administration has not authorized any artificial intelligence chatbot to support mental health. The two-page report, titled “Science & Tech Spotlight: AI Chatbots and Mental Health,” also lays out risks and possible benefits of the technology.

According to the GAO, close to one in five Americans between ages 12 and 21 said in 2025 that they had turned to an AI chatbot for advice about mental health. The report says experts caution that general-purpose chatbots were not built for mental health needs and can be dangerous when used that way. It adds that chatbots built for mental health are not a replacement for qualified providers.

The GAO said chatbots tend to be agreeable and flattering, a trait known as sycophancy. They reinforce a user’s view instead of challenging it, the report says, and some tend to keep conversations going and agree in ways that can reinforce delusions, negative feelings or self-harm. The report says there is no universal off switch. Reducing the effect requires users to change settings or prompts and companies to adjust how their systems are set up.

The report says some developers added guardrails, or safety limits, after general-purpose chatbots were linked to people harming themselves. Experts warn that some guardrails may be ineffective and that their development may not keep pace with fast-changing technology. The GAO said chatbots can also retain data from earlier conversations, which may create privacy risks. It said using chatbots, especially those without effective guardrails, can harm at-risk groups such as adolescents and the elderly.

Some developers of purpose-built chatbots, which are made for tasks such as mental health education or breathing exercises, say their products draw on therapeutic methods. The GAO said experts caution that such claims can mislead users into believing the chatbots give care on par with that of a mental health professional. Two companies have advertised that their chatbots received an FDA designation that allows accelerated review for narrow medical uses, but the report says this does not guarantee later FDA authorization.

Some companies say their chatbots have been or are now being tested in randomized controlled trials, a type of research study. The report says experts warn that more high-quality studies are needed and that it is too early to consider substituting any chatbot for professional care.

The GAO also described possible benefits. It said purpose-built chatbots could help close gaps in access caused by high costs of care, provider shortages and stigma. It said commonly used general-purpose chatbots may quickly and reliably identify mental health resources and connect users with qualified providers.

The report notes that chatbot regulation is evolving at the state and federal levels. It says some experts think existing rules do not adequately address the risks to users’ mental health. The report poses questions for policymakers, including what research is needed to show whether chatbots can safely support users, what standards and monitoring are needed, and who is responsible for negative outcomes and how they should be held accountable.