Stephanie Palazzolo reported that OpenAI's Astra AI uses recurrent depth as a reasoning method. The post notes the approach can aid model costs and performance. Several AI researchers and commentators reacted with concerns that the technique obscures the model's thinking process and reduces access to monitorable chain of thought. Posts reference prior OpenAI statements against similar methods and discuss whether recurrence necessarily limits interpretability. Joshua Achiam commented that chain of thought interpretability was always fragile for long-term safety. The account links the details to coverage in The Information.