OpenAI locks down Astra after model raises first-ever critical cyber capability fears
Interesting Engineering
OpenAI has classified its upcoming Astra model under its highest cybersecurity risk category after internal testing revealed advanced offensive cyber capabilities.Summary
OpenAI has classified its upcoming AI model, Astra, under its highest cybersecurity risk category after internal evaluations indicated it could possess advanced offensive cyber capabilities, potentially meeting the "Critical" threshold of its Preparedness Framework. This assessment has prompted the company to tighten internal security, introduce stricter protections in the development environment, and pause non-compliant work. OpenAI plans to collaborate with government agencies and independent safety groups for external testing and validation before deploying the model broadly to help strengthen digital defenses.
(Source：Interesting Engineering)