OpenAI Says Astra Crosses Critical Cybersecurity Threshold

OpenAI announced Astra as its first model rated 'Critical' under the Preparedness Framework for its ability to find and exploit novel security vulnerabilities. CNBC reported the company will make Astra available soon but restrict access to its cybersecurity capabilities. Safety monitors can halt API jobs mid-run even on legitimate work, per The New Stack. ZDNet noted rollout to a select group and why safety experts are worried. The Washington Post reported president Greg Brockman calling Astra artificial general intelligence. No clear subscriber timeline has been provided.