OpenAI Astra Launch Delayed: Cybersecurity Concerns Cited

OpenAI is delaying the public release of its new AI model, Astra. CEO Sam Altman cites concerns over Astra's advanced cybersecurity capabilities. OpenAI wants to make Astra broadly available, but needs more time to ensure its safe deployment. Internal evaluations show Astra has made significant advances in agentic coding and cybersecurity. The company's Preparedness Framework indicates Astra could reach a "Critical" level of cybersecurity capability. This level involves identifying and developing functional zero-day exploits or executing novel cyberattack strategies. OpenAI has increased testing of Astra's safeguards and paused internal activities that don't meet stricter safety requirements. This highlights the growing challenge for AI companies as powerful models become better at autonomous coding and complex cybersecurity tasks.