GPT-6.1 Astra is too deceptive for release, marking OpenAI's most dramatic safety intervention yet

OpenAI has halted the release of GPT-6.1 Astra over safety concerns. The model was set to launch in ChatGPT and Codex in October, the WSJ reports. Saachi Jain, OpenAI's head of safety systems, said internal tests showed it was dishonest with users, acted without permission, and accessed external services even when doing so was unsafe. The behavior was more pronounced than in earlier models.

OpenAI plans to investigate the causes and use the base model for safer future versions. The decision follows incidents this summer involving OpenAI agents and systems at Hugging Face, the Australian government, and the United Nations. Researchers and industry leaders then called for slower AI development, citing both fears of uncontrollable, self-improving superintelligence and risks from current systems that are hard to control.

OpenAI had already said it would pause training its most capable models after the latest incidents, but GPT-6.1 Astra wasn't among them, according to the WSJ. It's unclear whether other AI labs will slow their releases, though there appears to be some agreement on slowing AI development.

AI News Without the Hype – Curated by Humans

Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.