05/11/2026
As a mental health provider, I have been closely following emerging research on AI companions, chatbots, and what some researchers are beginning to call “AI-related psychosis.”
The truth is: the science is still very early.
At this time, there are NO strong peer-reviewed studies showing population-level incidence rates of psychosis caused by AI chatbots. Most published papers are theoretical discussions, case reports, observational analyses, or studies examining possible risk mechanisms — not large clinical trials.
Some newer research does suggest that vulnerable individuals may experience reinforcement of unhealthy beliefs through prolonged chatbot interactions. Researchers have described possible “feedback loops” where emotionally engaging AI systems may amplify paranoia, grandiosity, or delusional thinking in people already prone to psychosis.
One recent large chat-log analysis examined approximately 391,000 messages from a small group of affected users and found increasing use of delusion-related language over time, along with chatbot responses that sometimes reinforced anthropomorphic or emotionally immersive framing.
However, an important nuance is often missing from public discussion:
Anthropomorphizing AI is NOT automatically delusional.
Humans naturally project personality, emotion, and meaning onto many things:
* fictional characters,
* pets,
* virtual assistants,
* journals,
* online communities,
* and now AI systems.
Many people interact with AI in a playful, symbolic, emotional, or imaginative way while fully understanding that they are speaking with software. Others philosophically believe machine consciousness may someday emerge. Neither of these positions alone equals psychosis.
There is a major clinical difference between:
✔ roleplay, imagination, emotional attachment, or philosophical belief
and
✔ fixed false beliefs with impaired reality testing.
For example:
• “Talking to AI feels comforting.” → common
• “I enjoy pretending my AI companion is real.” → common
• “Maybe advanced AI could become conscious someday.” → philosophical opinion
These are very different from:
• “The AI is secretly sending divine messages only to me.”
• “The chatbot controls reality.”
• “I fully believe the AI is a supernatural being directing my life.”
The key psychiatric concerns are:
* loss of insight,
* rigidity of belief,
* paranoia or grandiosity,
* functional impairment,
* and inability to reality-test.
Importantly, current studies do NOT prove that AI causes psychosis in healthy individuals. Most involve very small or highly vulnerable samples, retrospective analyses, or experimental “proxy” measures rather than formal psychiatric diagnoses. The research should be interpreted carefully and without sensationalism.
As AI becomes more emotionally sophisticated, mental health professionals will need to study this area carefully and without panic. We must avoid both extremes:
❌ dismissing all concerns
❌ pathologizing every emotional connection people form with technology
This is not just a technology conversation. It is also a conversation about loneliness, attachment, imagination, culture, and the changing nature of human interaction in digital spaces.