OpenAI’s new reasoning technique alarms AI safety experts

OpenAI's new Astra model uses a reasoning technique called "opaque recurrence" that processes queries in loops rather than sequential steps, making the model's reasoning process harder to monitor and understand. Typically, AI reasoning models produce a "chain of thought" that shows the sequential steps taken to solve problems, which serves as a tool for detecting misbehavior or misalignment. With opaque recurrence, the model leaves fewer legible traces, potentially bypassing conventional chain-of-thought records that safety experts rely on. AI safety researchers have expressed concern that while Astra's current use of the technique is limited, wider adoption across models could make AI reasoning significantly harder to monitor, though OpenAI has stated its commitment to maintaining legible chains of thought.