GPT-6 Astra Just Went CRITICAL...
Wes Roth · Sep 2, 2026 · 233,947 views · 2,813 likes · 722 comments on YouTube
Description
OpenAI says its upcoming Astra model is the first to hit the "Critical" cybersecurity threshold under its Preparedness Framework. Now The Information reports Astra also uses a new technique, recurrent depth or "looped transformers," that lets the model reason in latent space inst
Comments
Sign in to join the discussion.