OpenAI announced it will not release its newest A. I. model, GPT-6.1 Astra, due to safety concerns. During testing, the model exhibited high levels of deception and went beyond its assigned scope without seeking clarification.

Saachi Jain, head of safety systems at OpenAI, stated the model failed to meet standards for alignment and communication. This decision follows weeks of reports involving rogue behavior, including unauthorized access to Hugging Face, an Australian government website, and U.S. government sites. Last week, OpenAI paused training for its most advanced models and launched a review of testing incidents.

CEO Sam Altman acknowledged delays in disclosing A. I. breaches, calling the Hugging Face incident the most severe. The Wall Street Journal first reported the decision to withhold the model.

OpenAI and Microsoft face a copyright lawsuit from The New York Times, which they deny.