OpenAI's Next AI Model Astra Shows Cyber Performance Strong Enough to Trigger Pause
Covered by 2 sources
Read full postOpenAI has paused some internal activities involving its upcoming AI model Astra after evaluations showed it may have critical cybersecurity capabilities, including developing zero-day exploits autonomously. The company is implementing enhanced security measures and collaborating with government and safety organizations to safely test Astra's capabilities.

Covered by 2 sources
- Further Developments About Internal AI Models Hacking Things· Don't Worry About the Vase
- OpenAI Astra pause 🚨, Claude Code cross-session 🤖, how Cursor Router works 🔀· 2 sources
- The AI safety test is becoming a safety risk· TechCrunch
- Now we have a timeline of the OpenAI accidental attack against Hugging Face· 2 sources
- Why Aren’t Any AI Companies Watching Their Frontier Models to Make Sure They Don’t Go on Hacking Sprees?· Futurism
- Why are so many AI models going 'rogue'? The experts weigh in · TechRadar


