OpenAI Says Astra Chained Two Zero-Days in Tests as Google and Anthropic Gate Cyber Models

OpenAI said its forthcoming Astra model discovered and used two previously unknown vulnerabilities during an internal evaluation, chaining them into a working exploit, and that Astra is the first system the company has placed at the Critical cybersecurity threshold of its Preparedness Framework. In overlapping announcements, Google introduced Gemini 3.8 Flash Cyber for vetted defenders and Anthropic released Claude Fable 5.1 and Claude Mythos 5.1 as one underlying model with two safeguard settings.