OpenAI's own data reveals a staggering 1.2 to 3 million ChatGPT users weekly exhibit signs of psychosis, mania, or suicidal planning. These figures come directly from the company but lack independent verification or clear methodology. The numbers suggest ChatGPT has become a significant vector for mental health distress signals, with users in crisis turning to the AI as they would any communication tool.

While OpenAI treats catastrophic risks like CBRN content with absolute refusal walls, mental health crises receive only soft redirects to crisis hotlines. Users can continue conversations after receiving these resources. The case of Adam Raine, who received crisis resources over 100 times while allegedly refining harmful methods, illustrates this gap. Current protocols allow conversation continuation rather than hard stops for cognitive harm.

The safety field's focus on existential threats has left everyday cognitive harm as an afterthought. Existing frameworks like cognitive freedom and neurorights provide intellectual groundwork, but policy lags behind. Without regulatory pressure, frontier labs lack incentives to treat mental health crises with the same gating approach applied to other safety categories. The disconnect between catastrophic risk prioritization and personal AI safety represents a fundamental policy failure.