OpenAI's new Astra model will use a reasoning technique called “recurrent depth,” also known as “opaque recurrence,” which allows it to operate outside of the sequential thinking typical of most reasoning models. This technique, first reported by The Information on Tuesday, has raised concerns among AI safety experts, who fear it will make the model’s chain of thought more difficult to monitor. Redwood CEO Buck Shlegeris expressed extreme concern, stating that pushing this technique further could "totally destroy CoT monitorability." AI safety advocate Zvi Mowshowitz warned it was "playing with fire." OpenAI chief scientist Jakub Pachocki emphasized the lab's commitment to legible chains of thought, and the company stated Astra's use of the technique is currently limited. However, The Information also reported that Anthropic and Google DeepMind are already discussing the technique.