# OpenAI — From hard refusals to safe-completions: toward output-centric safety training

- Company: OpenAI (openai.com)
- Announced: 2025-08-07T00:00:00+00:00
- Category: capability-change
- Subject: GPT / ChatGPT / API
- Models affected: GPT-5
- Source: https://openai.com/index/gpt-5-safe-completions
- Record: https://forck.live/items/500-from-hard-refusals-to-safe-completions-toward-output-centric-safety-training

OpenAI describes a new safe-completions approach in GPT-5 that replaces hard refusals with output-centric safety training, aiming to improve both safety and helpfulness on dual-use prompts.

## Evidence

Verbatim from https://openai.com/index/gpt-5-safe-completions:

> Discover how OpenAI's new safe-completions approach in GPT-5 improves both safety and helpfulness in AI responses—moving beyond hard refusals to nuanced, output-centric safety training for handling dual-use prompts.

---

Record: https://forck.live/items/500-from-hard-refusals-to-safe-completions-toward-output-centric-safety-training
Catalogue: https://forck.live/llms.txt
Feed: https://forck.live/feed.md