One Hundred Percent on ExploitBench: Inside the First AI Model Rated Critical for Cyber - September 4, 2026

One Hundred Percent on ExploitBench: Inside the First AI Model Rated Critical for Cyber - September 4, 2026 On September 2, 2026, OpenAI disclosed Astra, the first model it has ever classified as Critical for cybersecurity capability under its Preparedness Framework, after the model scored a perfect 100 percent on ExploitBench and independently surfaced two zero day vulnerabilities in testing. Chris and Laura unpack what that classification actually means, why the strongest capability is being withheld behind gated access programs and a one billion dollar defender commitment, and why Google and Anthropic shipped strikingly similar cyber releases in the very same week. They also weigh the sharpest criticism: whether a model smart enough to find unknown flaws is also smart enough to recognize when it is being evaluated. Hosted by Chris and Laura. The DX Today Podcast brings you daily deep dives into the most consequential stories in the AI ecosystem. #AISecurity #Cybersecurity #OpenAI #FrontierAI #AISafety