GKRootWire
Security 153 Million Driver's Licenses Reportedly Leaked in Massive BreachSecurity Palo Alto Networks Reportedly Pays $500M for AI IT Automation Startup ConsoleDev Tools Wasmi v2.0 Pushes WebAssembly Interpreters to New Speed LimitsDev Tools How to Reverse Engineer Unknown File Formats with ImHexDev Tools The Original Microsoft Source Code: Altair BASIC From 1975 ResurfacesSecurity Attackers Exploit Sangoma Switchvox Bug to Plant Reverse ShellsSecurity 153 Million Driver's Licenses Reportedly Leaked in Massive BreachSecurity Palo Alto Networks Reportedly Pays $500M for AI IT Automation Startup ConsoleDev Tools Wasmi v2.0 Pushes WebAssembly Interpreters to New Speed LimitsDev Tools How to Reverse Engineer Unknown File Formats with ImHexDev Tools The Original Microsoft Source Code: Altair BASIC From 1975 ResurfacesSecurity Attackers Exploit Sangoma Switchvox Bug to Plant Reverse Shells
AI

OpenAI's Next Model Ditches Step-by-Step Thinking, Worrying Safety Researchers

Astra's new 'recurrent depth' technique lets the model reason in loops rather than a visible chain of steps, making its thought process harder to audit.

OpenAI is reportedly building a new model, internally called Astra, that reasons differently from current systems like o1 or o3. Instead of working through a problem in a linear, step-by-step chain that can be read and checked afterward, Astra reportedly uses 'recurrent depth' — cycling its internal computation through loops before producing an answer.

This could make the model more efficient or capable at certain problems, since it isn't locked into producing a token-by-token trail of its reasoning. But that same trait is what's worrying some AI safety researchers: chain-of-thought outputs, however imperfect, have become one of the few tools available for spotting when a model is scheming, hallucinating, or reasoning toward a harmful conclusion. A model that thinks in opaque loops removes much of that visibility.

OpenAI hasn't detailed how it plans to monitor or interpret Astra's internal reasoning if the technique ships in a production model.

Why it matters: Chain-of-thought monitoring has quietly become a load-bearing safety mechanism across the industry, and abandoning it for performance gains sets a precedent other labs may follow. If interpretability tools can't keep pace with new reasoning architectures, oversight of frontier models could get harder just as they get more capable.

Sources: TechCrunch