Some Codex users say their opt-out from model training switched back on

A Hacker News post on 3 November says the "Improve the model for everyone" setting had been re-enabled in Codex for the poster and coworkers, though none of them knowingly opted back in; their company had turned off data sharing to protect its IP. The poster points to chatgpt.com/settings/data-controls to check the setting, and to OpenAI's privacy portal as another way to turn data sharing off. These are user reports, not confirmed by OpenAI when we read the thread. Among 41 top-level replies, four people said the same had happened to them and five said their setting was still off; two suggested a bug, while four said firms that care should use a zero-data-retention contract, a cloud provider or their own models instead. No consensus on the cause. Sorted take: a settings toggle is not a contract, so check it and keep a record.