Saved report copy
Recorded
OpenAI’s Astra reaches critical cyber threshold, raising dual-use oversight concerns
WIRED reports that OpenAI’s forthcoming Astra model can independently find and exploit unknown software vulnerabilities, including by chaining exploits, prompting a pause in development for added safeguards. OpenAI plans limited early access for selected cybersecurity partners while restricting advanced capabilities for most users, highlighting the need for oversight of offensive cyber tools.