Helping ChatGPT better recognize context in sensitive conversations

OpenAI has introduced new safety updates for ChatGPT aimed at improving context awareness during sensitive conversations involving potential risks like suicide, self-harm, and harm-to-others. According to OpenAI, the updates enable the AI to better recognize when risk emerges over time by identifying subtle or evolving cues within single conversations or across separate interactions. To help ChatGPT connect relevant signals without overreacting in everyday chats, the company developed safety summaries, which are short, factual notes about earlier safety-relevant context created by a specialized model. These summaries are narrowly scoped, retained for a limited time, and utilized solely for serious safety concerns. The updates build on more than two years of collaboration with mental health and safety experts, including psychiatrists and psychologists from OpenAI's Global Physicians Network. Internal evaluations demonstrated significant improvements in safe-response performance, with increases of up to 50% in long single-conversation suicide and self-harm scenarios, and up to 52% in multi-conversation harm-to-others scenarios using the GPT-5.5 Instant model. OpenAI also evaluated the safety summaries across more than 4,000 assessments, yielding high average scores for safety relevance and factuality, while finding that ordinary conversation quality remained comparable. OpenAI stated it may explore applying similar methods to other high-risk areas like biology or cyber safety in the future.