OpenAI seems to be silently re-enabling model training on Codex chats

OpenAI appears to have silently re-enabled model training on Codex chats, with users reporting that the "Improve the model for everyone" data-sharing setting was switched back on without their knowledge. The issue, flagged by employees at a company concerned about leaking intellectual property through prompts, was noted previously on Hacker News (item 49643556). Affected users can review their status at chatgpt.com/settings/data-controls or opt out via privacy.openai.com. The apparent reset matters because enterprises increasingly route proprietary code through AI coding tools and rely on opt-out controls to protect trade secrets. What to watch: whether OpenAI confirms a bug or policy change, how broadly settings were affected, and whether enterprise customers demand contractual guarantees or audit logs for data-sharing consent.