OpenAI Reportedly Flipped Your Codex Data Sharing Setting Back On

OpenAI appears to have silently re-enabled data sharing for some Codex users, a move that breaks trust and puts proprietary code at risk. If you have ever used Codex and opted out of data sharing, you need to go check your settings again right now. Your choice may have been reverted without your consent.

What happened?

Multiple reports on Hacker News surfaced from developers who discovered the "Improve the model for everyone" toggle in their OpenAI settings was active. These users were certain they had previously disabled it to protect their intellectual property. The change happened silently, with no notification or explanation from OpenAI.

What's the real IP risk?

For anyone building with Codex, allowing OpenAI to use your data for training creates a direct pipeline for your source code into their models. This means proprietary algorithms, business logic, and internal comments could be absorbed and potentially regurgitated in code suggestions for other users, including your competitors. The opt-out is the only safeguard, and if that can be undone without notice, the service is not safe for commercial development.

What should you do now?

Log in to your OpenAI account and navigate to the settings page. Find the data controls and verify the status of the toggle for model improvement. Do not assume your previous opt-out is still valid. Take a screenshot for your records. This incident shows that consent settings for AI platforms require periodic auditing, not just a one-time setup.

The verdict

The silent reversal of an explicit user choice is a major failure. Without a transparent post-mortem from OpenAI and a technical guarantee that user settings will be respected, I would not use Codex for any project involving proprietary code. The risk of IP leakage is too high, and the trust is broken. Use tools that guarantee data privacy through their business model, not just a toggle in a settings menu.

FAQ

Does this data sharing issue affect the OpenAI API? Data submitted via the OpenAI API has not been used for training since March 2023, according to their policies. This issue appears specific to user interfaces like the Codex playground, but the lack of communication erodes trust across all services.

What's the specific risk of my code being used in training? If your code is used for training, it could become part of a future model's knowledge base. This means fragments of your proprietary logic or entire functions might be suggested to other users, potentially exposing trade secrets or internal architecture.

Has OpenAI commented on this settings reversal? As of this writing, OpenAI has not released a public statement or explanation for why these user data sharing settings were reverted. The discovery was driven entirely by users auditing their own accounts.