OpenAI Delays Astra: Cyberattack Risk Halts AI Model

OpenAI is delaying the release of its advanced AI model, Astra. The company states that internal testing showed the system could potentially plan and execute a cyberattack independently. This marks the first time a major AI lab has publicly slowed a model due to cybersecurity concerns. OpenAI's internal evaluations found Astra made significant progress in agentic coding and cybersecurity. This led them to believe Astra could reach the "Critical" tier in their Preparedness Framework. A "Critical" rating means a model could autonomously discover and build zero-day exploits, or independently design and carry out a new, end-to-end cyberattack strategy. OpenAI has suspended parts of Astra's development that do not meet new security standards. They have also implemented a broader monitoring system for the model. This pause highlights growing concerns about the autonomous capabilities of advanced AI.