Addendum to GPT-5 System Card: Sensitive conversations

Source: OpenAI Date: 2025-10-27 URL: https://openai.com/index/gpt-5-system-card-sensitive-conversations

Summary

System card addendum for GPT-5 covering sensitive conversation handling — mental health, self-harm, crisis situations, political content, and other high-stakes dialogue topics. The addendum documents specific evaluation results for these domains and explains the trade-offs OpenAI made between helpfulness and harm avoidance in each category.

Implications

Safety/policy thread. Sensitive conversation addenda (see also the 4o image generation addendum) are becoming a standard pattern in OpenAI’s governance documentation — acknowledging that specific topic areas require safety accounting beyond the base system card. The mental health section is particularly significant given the October 2025 teen safety blueprint and the expert council on well-being — OpenAI was under scrutiny for how GPT-5 handles vulnerable users at this time.