OpenAI pauses Astra work after model nears critical cyber threshold
What's new: OpenAI moved Astra into isolated test environments and restricted network access while outside experts assess its hacking ability.
OpenAI halted some internal work on its upcoming AI model Astra after early testing showed it may have reached a critical cybersecurity threshold. The company said recent internal evaluations and outside expert reviews suggested Astra could autonomously carry out increasingly sophisticated cyber tasks, including potentially finding and exploiting severe software flaws. In response, OpenAI tightened security rules, suspended Astra-related work that does not meet the higher standard, and shifted development into isolated testing environments with restricted network access and sandboxed execution. OpenAI said Astra was not involved in the July hack targeting Hugging Face and that it will work with government agencies and selected AI safety groups to further test the model.