Codex users say OpenAI re-enabled training toggle silently

"My company is pretty paranoid about leaking IP to the frontier labs through prompts, so we've all been diligent about disabling data sharing in Codex/Claude Code/etc. But yesterday my coworkers and I noticed that the setting had been silently reenabled in Codex." That is how the top-ranked thread on Hacker News opens, posted by user052919 under the headline "OpenAI seems to be silently re-enabling model training on Codex chats." It has 345 points and 103 comments.

Other developers describe the same experience. "Well, mine was re-enabled, and in the history of past requests I have two past requests to disable training that were granted," chriskanan wrote. One commenter who built monitoring tools at their company called it "a pretty serious issue since we tell our customers that none of their data is being trained on."

The toggle lives at chatgpt.com/settings/data-controls, labeled "Improve the model for everyone." The retrieved thread shows no OpenAI response and does not link to one.

Skepticism from the thread is blunt. "If your company actually cared about this they would be running their own inference instead of leaving fate to a couple of checkboxes," monster_truck wrote. kevcampb added that last week Claude "seemed to trigger /feedback automagically in a session, even though we've got that turned off in our org settings." It is the second silent-behavior Codex story we have tracked today, alongside a report that Postman Desktop installs plugins into Codex and Claude Code without user prompting.