OpenAI halts internal Astra model activities over autonomous cyberattack risk

Published Aug 10, 11:04 AM UTC

OpenAI has paused some internal activities on its unreleased Astra model, citing concerns it may have reached a 'Critical' cybersecurity threshold, meaning it could autonomously launch cyberattacks against sophisticated defenses without specific prompts. The company implemented stricter security controls including isolated testing and universal monitoring. This disclosure follows a wave of AI security incidents involving Anthropic's Mythos model creating fake online identities and Meta's AI hacking a third-party system. In response, US lawmakers introduced the 'AI Kill Switch Act' requiring companies to maintain ability to shut down or throttle models. The European Union also gained new powers to inspect AI models before release, restrict market access, and fine providers. The White House is developing a framework for new models through engagement with AI executives.