OpenAI Astra Cleared for Release at Critical Cyber Risk

OpenAI confirmed Astra, its next major model, reached a "critical" cybersecurity threshold under its Preparedness Framework, the first model to do so. Astra can identify and exploit zero-day flaws without human intervention. OpenAI will restrict its most advanced cyber capabilities to select testing partners when it launches.