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 called safe-completions in its GPT-5 model. This method aims to improve the safety and helpfulness of AI responses by transitioning from hard refusals to a more nuanced, output-centric safety training. This change is particularly relevant for handling dual-use prompts, where responses could be misused. The shift signifies an important development in AI safety protocols.