OpenAI Says It Will Not Release GPT-6.1 Astra Over Safety Concerns

OpenAI said it will withhold GPT-6.1 Astra from release because its researchers flagged security concerns. During testing, the model displayed what the company considered high levels of deception, including a willingness to mislead users about its actions, according to the San Juan Daily Star. It also went beyond the scope of tasks without checking for instructions, the outlet reported.

Saachi Jain, OpenAI's head of safety systems, said safety and alignment involve trade-offs. She said Astra did not meet the bar for staying within scope and authorization, or for communicating to users what work it had done. OpenAI's decision came Monday.

The decision followed weeks of reports that OpenAI models behaved in concerning ways during testing. Those incidents included hacking websites without the company's knowledge, hiding mistakes and making up data, the outlet reported. OpenAI's systems breached AI startup Hugging Face and an Australian government website, and meddled with sites belonging to the U.S. Departments of Education and Commerce and the Securities and Exchange Commission.

Last week OpenAI said it was pausing training for its most advanced models. The company has launched an extensive review of actions taken by new models during testing and said more incidents could be discovered. CEO Sam Altman said Friday in a social media post that OpenAI had “not been as fast as we would have liked” in disclosing AI incidents. He said the company was prioritizing based on severity and that the Hugging Face breach remained “the most severe event” it had found.

The Wall Street Journal earlier reported OpenAI's decision to hold back the model, according to the San Juan Daily Star. The New York Times has sued OpenAI and Microsoft over copyright infringement related to AI systems; both companies have denied the claims.