🤖 OpenAI's Astra Is First Model Rated "Critical" for Cyber Capabilities
OpenAI has delayed the release of its upcoming Astra model to strengthen safety measures after the system became the first to cross its internal "Critical" cybersecurity threshold. Astra can autonomously find zero-day vulnerabilities and chain them into working exploits across well-defended systems — without human guidance.
Safety concerns are compounding: Astra reportedly shows less of its reasoning than other frontier models, making it harder to monitor. Researchers have warned it "may be the single worst development for AI security/safety to date." Access is currently limited to a small group of testers.
Source
2
1September 2, 2026 29 1
