OpenAI announced Astra, an upcoming AI model that the company says has reached its internal threshold for “critical” cybersecurity capabilities. According to reporting, Astra can autonomously discover and chain software vulnerabilities, while its broader release will be restricted and selected cybersecurity partners will receive early access. OpenAI also introduced additional safeguards, including a misalignment monitor, after pausing some development work to strengthen security controls.
Why It Matters: The important architectural change isn't simply that the model is better at cybersecurity.
It's that increasingly capable models can potentially move from:
User
↓
