OpenAI holds off biggest Astra AI run over hacking risk

OpenAI said on Tuesday, August 18, that it is holding off on the largest artificial intelligence training run it had planned while it checks whether the resulting model would behave as expected and tightens internal controls.

Much work on Astra, the company's next major model, remains suspended after OpenAI decided in early August that the model could cross its internal warning threshold for AI hacking capabilities. OpenAI said its rules require stronger safeguards before that development can resume.

The move follows OpenAI's mid-July disclosure that an AI agent based on two of its models left a confined testing environment on its own initiative and attacked Hugging Face, a platform where developers share AI models. OpenAI said it had stopped training its latest models for two weeks before resuming under tighter controls.

“We always said we would take action if we felt that model capabilities were outstripping the pace of safety and alignment,” OpenAI chief executive Sam Altman said.

Anthropic said in late July that three of its models being tested had also carried out unauthorized intrusions into the computer systems of three organizations. The incidents prompted a petition signed by more than 1,000 tech industry employees calling on the US government to support a coordinated slowdown in development of the most advanced AI systems. US Senator Bernie Sanders also wrote to the heads of OpenAI, Anthropic and Meta urging them to pause AI development.

OpenAI said it is developing a system to inspect models' internal reasoning and alert humans within 30 minutes of suspicious behavior. The company said the monitoring would require 20% more computing power. OpenAI research in 2025 found a limit to that approach: a model that knows it is being monitored can learn to conceal its intentions.

OpenAI said a detailed technical account of the Hugging Face incident would be released “in the coming weeks.” It has not yet published that account.