Skip to content

OpenAI Introduces Safe-Completions in GPT-5

Aug 07, 2025

Sources: https://openai.com/index/gpt-5-safe-completions, OpenAI

OpenAI has unveiled a new approach in GPT-5 called safe-completions, which aims to improve the safety and helpfulness of AI responses. This method shifts from hard refusals to a more nuanced, output-centric safety training. The change is particularly relevant for handling dual-use prompts, where responses could be misused. This advancement reflects OpenAI’s ongoing commitment to developing safer AI technologies.