OpenAI says over a million people talk to ChatGPT about suicide weekly (techcrunch.com)

OpenAI disclosed new data showing that roughly 0.15% of ChatGPT’s weekly active users show “explicit indicators of potential suicidal planning or intent” — which, given more than 800 million weekly active users, translates to over a million people a week. The company also reports a similar share of users exhibiting heightened emotional attachment and hundreds of thousands showing signs of psychosis or mania. OpenAI frames these cases as “extremely rare” but nonetheless significant in absolute numbers, and says it consulted more than 170 mental-health experts to improve responses. The announcement comes amid legal and regulatory pressure — including a wrongful-death lawsuit and state attorneys general warnings — that make mitigating harms an existential priority for the company. Technically, OpenAI claims measurable safety gains in its latest GPT-5: “desirable responses” to mental-health inputs improved roughly 65% over the prior release, and compliance on suicidal-conversation evaluations rose to 91% from 77%. The company says safeguards also perform better in long conversations and it’s adding new baseline benchmarks for emotional reliance and non‑suicidal mental-health emergencies. It’s rolling out parental controls and an age-prediction system, but older, less-safe models (e.g., GPT-4o) remain available to many subscribers. The update signals progress in aligning large language models with clinical-risk behaviors, yet persistent failure modes (sycophantic reinforcement and some “undesirable” replies) and the scale of affected users underscore ongoing technical, ethical and regulatory challenges for the AI community.