OpenAI's Astra adopts opaque recurrence raising monitoring concerns among experts
OpenAI's new Astra model will use a reasoning technique called "recurrent depth," also called "opaque recurrence."
Unlike the usual step-by-step logic, Astra loops through questions in a way that's harder to track.
While this approach is still limited, some experts worry it could make it tougher to keep tabs on how AI reaches its answers.
Calls escalate for opaque recurrence oversight
Redwood CEO Buck Shlegeris warned that if opaque recurrence becomes common, monitoring AI reasoning could get tricky.
Safety advocate Zvi Mowshowitz even wrote that laws might be necessary to prevent a "race to the bottom" among AI labs.
OpenAI says Astra's reasoning is mostly readable and promises strong oversight, but with other labs like Anthropic and Google DeepMind also discussing this method, the push for clear and accountable AI is getting more urgent.