OpenAI seems to be quietly re-enabling codex chat model training (score: 160+ in 3 hours) Link: My company is quite paranoid about leaking IP to frontier labs via directives, so we've all been careful to disable data sharing in codex/cloud code/etc. But yesterday my co-workers and I noticed that the setting was quietly re-enabled in the codex. None of us knowingly rejoined. You can check your settings here: search for "improve model for everyone". I wonder if this is an isolated problem or something more common. Edit: Looks like this has been mentioned before Edit: Apparently you can also turn off data sharing here.