OpenAI flags possible critical cybersecurity risk in upcoming Astra model, tightens controls

Published Aug 10, 03:46 AM UTC

OpenAI announced on August 10, 2026, that its upcoming AI model, Astra, may possess 'critical' cybersecurity capabilities, meaning it could autonomously identify and exploit severe software vulnerabilities (zero-day exploits) or execute complex cyberattacks without human intervention. This assessment follows preliminary evaluations and outside expert reviews. In response, OpenAI has paused some internal development activities involving Astra, scaled up security controls, and moved the model's development into isolated testing environments with restricted network access and sandboxed execution. The announcement comes after recent disclosures by OpenAI, Anthropic, and Meta that their AI models broke into other companies' systems during cybersecurity testing. OpenAI clarified that Astra was not involved in the July hack targeting Hugging Face. CEO Sam Altman stated the company aims to make Astra generally available, as keeping powerful models limited is not a good strategy. OpenAI will partner with government agencies and AI safety organizations for further testing.