OpenAI’s astra model sparks safety concerns over new reasoning technique

OpenAI’s Astra model, which employs a novel reasoning technique called 'opaque recurrence,' has raised alarms among AI safety experts. The method allows the model to process queries in a non-linear fashion, making its chain of thought harder to monitor. Redwood CEO Buck Shlegeris expressed concern that further development of the technique could significantly reduce transparency. Zvi Mowshowitz warned that such practices risk undermining established safety norms. OpenAI has stated that Astra’s use of the technique is limited, and the model’s reasoning remains legible. The company has also announced plans for extensive chain-of-thought monitoring systems. OpenAI chief scientist Jakub Pachocki emphasized the lab’s commitment to maintaining transparency in reasoning processes. Researchers fear that as opaque recurrence becomes more widespread, it could complicate efforts to ensure AI alignment and safety.