product1 distinct publisher
OpenAI's GPT-6 Astra pairs harder-to-monitor reasoning with a pledge to pause scaling if oversight slips
OpenAI's new model thinks repeatedly before it acts, and according to Manifold Security's CTO it usually does so without leaving the reasoning trace that agent audits read. Oversight moves to the buyer.
Publishers:techradar.com
Reality
- Evidence30
- Adoption18
- Hype gap+30
- Incentives82