OpenAI00:00PolicyOfficial Blog
ChatGPT improves context-aware safety detection
Safer replies in long chats, with clearer support routing.
Key Points
- 1Judges intent using surrounding context
- 2Focuses on de-escalation and support, not just refusal
- 3Officially explains what changed and why
OpenAI detailed safety updates that better detect risk emerging over time in a conversation. The system uses surrounding context to refuse, de-escalate, and route users to support. This reduces harmful escalation in sensitive chats. False positives may occur, so clarify intent if needed.