OpenAI's Astra Becomes the First Model to Hit Its 'Critical' Cybersecurity Threshold

OpenAI says its upcoming Astra model is the first to meet the Preparedness Framework's Critical cybersecurity capability threshold — able to find and chain zero-day exploits in hardened real-world systems — and detailed the safeguards it's deploying alongside it.