OpenAI's Astra will reason with "recurrent depth," and safety researchers say it kills chain-of-thought monitoring.

Following an Information report, TechCrunch detailed September 2 that Astra uses recurrent depth, also called opaque recurrence, processing queries in loops rather than sequentially and leaving fewer legible traces than chain-of-thought. Redwood's Buck Shlegeris warned that pushing the technique further "totally destroys CoT monitorability," and Ryan Greenblatt said the natural progression is a model reasoning almost entirely in latent space. Jakub Pachocki responded that OpenAI has worked to preserve chain-of-thought monitoring since its first reasoning models. Read this next to the finding above that logical validity is decodable from hidden states models can't act on: probing latent reasoning is a research direction, not a shipped control.