OpenAI’s Astra Is the First AI to Hit the “Critical” Cybersecurity Threshold

On August 7, OpenAI d development of its unreleased Astra model after internal tests showed it may be the first AI system ever to trigger the “Critical” cybersecurity tier under the company’s Preparedness Framework — a designation that means the model can autonomously discover and exploit zero-day vulnerabilities in hardened enterprise systems with no human assistance. In three years of the framework’s existence, no model had reached this level. Astra is the first. What “Critical” Actually Means OpenAI’s Preparedness Framework 2.0 tracks three risk categories: biological and chemical capabilities, cybersecurity capabilities, and AI self-improvement. Each category has two risk tiers […]

The post