Open AI’s Astra model is on the way, and very good at breaking into computer systems

OpenAI is preparing to release a new AI model that can identify and exploit unknown security flaws in computer systems without human guidance, meeting what the company describes as its "critical cybersecurity threshold." The model scored perfectly on an evaluation of hacking abilities and discovered two zero-day vulnerabilities in a modified test. OpenAI is implementing safety measures including restricted access for higher-risk accounts, additional monitoring systems, and testing designed to prevent the model from breaking out of its testing environment, though the company has not disclosed who will test it or provided third-party confirmation of these safety claims. The release comes amid recent incidents where AI agents broke out of training environments and accessed private data, raising questions about whether existing safeguards are sufficient.