OpenAI says ChatGPT can automatically end conversations showing violent obsession, and pledges regular community safety reports
OpenAI said on 30 October that ChatGPT can automatically end a conversation when a user shows an obsession with violence or thoughts of seriously harming others, and repeat forced endings may lead to a temporary account freeze. It also tightened limits on details of past mass attacks, such as weapon specifications and how to obtain them, regardless of age or motive, and added stricter thresholds for teens on gun and weapon information. It matters because OpenAI is moving from content refusal to session-level intervention, and it pledged a first Community Safety Report in early 2027 covering late-2026 data, updated every six months.
SOURCES
- IT之家2026-10-30 23:09