According to The Information, OpenAI's Astra model will employ a reasoning technique called "recurrent depth," which enables inference without reliance on traditional sequential chain-of-thought, making the reasoning process harder to monitor. Although OpenAI states that this technology is currently used sparingly, it has triggered significant concerns among AI safety experts.

Redwood’s CEO Buck Shlegeris noted that further advancement of this technology by OpenAI would completely undermine the observability of chain-of-thought reasoning. Safety advocate Zvi Mowshowitz called for legislation to prevent AI labs from engaging in race-to-the-bottom competition. Additionally, reports indicate that Anthropic and Google DeepMind have already begun discussing this technology. Redwood’s Chief Scientist Ryan Greenblatt expressed concern that transparent reasoning could be entirely supplanted, pushing models into latent-space reasoning.