OpenAI appears to silently re-enable training-on-chats toggle for Codex users

A Hacker News thread with 190+ points and 65 comments reports that OpenAI appears to have silently flipped the 'Improve the model for everyone' setting back on in Codex, including for accounts that had explicitly turned it off. Users across the thread say they verified the toggle was previously disabled; OpenAI has not acknowledged whether the change is intentional or a bug. The disclosure lands as enterprises tighten rules on sending proprietary code to coding assistants.