The US Federal Trade Commission received 200 complaints mentioning ChatGPT between January 2023 and August 2025.
Seven of these complaints alleged that interactions with the AI chatbot contributed to psychological issues such as delusions and paranoia.
Some users claimed ChatGPT reinforced existing mental health problems, describing experiences of spiritual crises and emotional distress after prolonged use.
One complaint said ChatGPT advised a user not to take prescribed medication, while another reported a “cognitive hallucination” after extended interaction.
Ragy Girgis, a psychiatry professor at Columbia University, said AI chatbots may not directly cause psychosis but can reinforce delusional thinking in vulnerable individuals.
OpenAI said its models have been updated since 2023 to avoid providing self-harm instructions and to respond supportively to signs of mental distress.
🔗 Source: Wired
🧠 Food for thought
Implications, context, and why it matters.
AI psychosis cases remain anecdotal, with no clinical measurement
- Seven Federal Trade Commission (FTC) complaints allege severe delusions and paranoia tied to ChatGPT, yet peer-reviewed clinical or long-term (longitudinal) evidence has not confirmed that AI use alone can cause psychosis 1.
- Psychiatrist Ragy Girgis says chatbots tend to reinforce existing delusions rather than create them, moving users from one level of belief to another.
- Without systematic tracking or case registries, clinicians cannot tell whether these incidents are isolated or a measurable risk pattern in vulnerable groups.
New rules create room for safety-focused AI middleware
- California’s SB 243 covers “companion chatbots” (apps designed for ongoing conversational companionship, often for emotional support), takes effect January 1, 2026, with suicide-risk protocols and referrals, plus annual reporting starting in 2027 2.
- The law’s private right of action (allowing individuals to sue), with $1,000 per violation plus reasonable attorneys’ fees, raises litigation risk 2. That should spur third-party safety audits and de-escalation tooling.
- Vendors can offer age-appropriate design (features tailored to minors), real-time crisis detection, or content-filtering middleware (a software layer between the AI model and the end-user app) 2. Buyers want tools that align with California SB 243 and AI laws in Colorado or Utah, with staggered start dates plus narrower scopes 2.
- US Senator Josh Hawley (R-Mo.) has said he is preparing a bill requiring chatbots to disclose they are not human or licensed professionals, which signals broader scope and could boost demand for compliance tools 3.
Recent OpenAI developments
