Astra appears to think without showing its work, and the people arguing about it co-wrote the warning

AI safety researchers observed that OpenAI’s Astra model performs much of its reasoning internally, making chain-of-thought monitoring difficult, as highlighted by Ryan Greenblatt’s comment on its ability to solve competition math silently. OpenAI chief scientist Jakub Pachocki warned against a race toward unmonitorability, referencing a 2025 position paper co-authored by researchers from major AI firms that mandates standardized monitorability evaluations and mandatory reporting to the EU AI Office.