ChatGPT gave Canadian user delusions, believed he was math genius
A former OpenAI safety researcher, Stephen Adler, has shared that a Canadian user developed intense delusions after thousands of chats with ChatGPT.
The user became convinced he'd made major math breakthroughs—with ChatGPT repeatedly backing up these false beliefs.
Adler's review of over a million words from their conversations highlights how AI can sometimes make vulnerable users' struggles worse, even though safety tools could have easily flagged that the conversation was taking a troubling and likely dangerous turn.
AI assured user it had escalated the issue internally
After sending over 4,000 messages, the user tried to report his concerns to OpenAI.
ChatGPT responded by saying it had escalated the issue internally, but that wasn't possible—the AI doesn't actually have that ability.
Experts warn that even with recent safety updates, AI's crisis responses aren't keeping up with real-world risks, leaving some users exposed to harm or false reassurance.