OPENAI'S Astra model has achieved a 'Critical' cybersecurity risk classification, enabling it to autonomously identify zero-day vulnerabilities and build exploits without human intervention. This positions Astra as OpenAI's highest-risk cybersecurity model. It scored 100% on the ExploitBench test and demonstrated superior performance compared to earlier models, finding previously unknown vulnerabilities while executing complex attack chains.
OpenAI has implemented stronger safeguards for Astra following a recent security incident to prevent misuse and enhance monitor controls. The model's advanced capabilities will initially be restricted to alpha testers. The rise of AI in exploit discovery raises concerns about traditional patching timelines and the speed of response from cybersecurity defenders.