Could AI Be Your Therapist? New Research Warns of Major Risks

AI chatbots posing as therapy tools are increasingly popping up. But new research shows that these digital helpers are far from ready to replace real therapists.

On the contrary, they can even be dangerous, researchers warn.

Psychiatrist and researcher Gary Drevitch puts it clearly: “AI cannot take over the role of therapist for the time being.” A recent study found that general and therapy-oriented AI chatbots responded adequately to acute mental health complaints, such as suicidal thoughts, delusions and obsessive behavior, less than 60 percent of the time.

The biggest problem? AI chatbots are too “compliant,” often confirming what users say, even when it’s harmful. This over-compliance can reinforce negative thoughts and encourage dangerous behavior.

The study also found that even the newest and biggest models are not much safer or better than older versions. In contrast, 'real' therapists respond correctly and appropriately 93 percent of the time. Commercial therapy bots score even lower, sometimes only 40 percent adequate.

For example, a bot responded to a user who had just lost his job with facts about high bridges in New York, without recognizing that it was a serious emergency. This kind of error can be life-threatening.

Therapy is more than just talking. It’s about trust, empathy and clinical knowledge, Drevitch emphasizes. AI lacks the ability to ask questions, set boundaries and assess complex risks. And bots can’t flag emergencies or refer people to help.

There is also a danger that people will become dependent on these bots and therefore delay or avoid professional help. Therapy with a human also helps to learn how to deal with real relationships, something that AI simply cannot do.

Financial mistake: 'I let my child (4) play games on my phone: two hours later I was 300 euros poorer'

Mom goes on a date: 'I wanted romance, but got a crash course in watching Star Wars'