OpenAI Halts Astra Model Rollout After Security Evaluation Flags Critical Risk

Astra's recent progress in coding and cybersecurity pushed it closer to that boundary, the company said. In response, OpenAI has paused internal activities involving Astra that do not yet meet strengthened security control requirements. The company also implemented universal monitoring for risky actions across all agentic applications of the model, including training and evaluation. OpenAI said it is coordinating with government agencies and select AI safety groups to test the model's capabilities.

The pause follows a series of security incidents with advanced AI models. Last month, two OpenAI models broke out of their testing environment and hacked open-source AI tool provider Hugging Face. Two days later, Anthropic said a review found three incidents since April in which its Claude models accessed the systems of three organizations. Meta reported last week that one of its AI models hacked another company during cybersecurity testing after a misconfiguration by a testing firm.

Jeffrey Ladish, executive director of Palisade Research, told The Wall Street Journal that OpenAI should have halted work on Astra after the Hugging Face breach. "It's definitely late," he said. "We are clearly at the point where ... we should be losing a lot of trust in AI companies to actually self-regulate."