OpenAI "Allow training" checkbox re‑enables itself – user reports and implications

The core finding

Users have observed that the "allow training" (or "Improve model for everyone") checkbox in ChatGPT’s data‑control settings can silently re‑enable itself after being turned off, even when users keep careful records of the change.

Why the behavior matters

  • Privacy consent – If the toggle does not reliably persist, users cannot be sure that their data is excluded from model training, contrary to OpenAI’s public statements.
  • Legal exposure – Inconsistent enforcement of the opt‑out could undermine any legal claim OpenAI might make about respecting user preferences.
  • Trust erosion – Re‑enabling a privacy‑related setting without user action resembles the “hidden‑default” tactics seen on other platforms and fuels broader skepticism about AI companies.

What users have reported

Observation Details Notable comments
Toggle flips back on Users who disabled the checkbox saw it re‑enabled on a later visit, sometimes after a product update or credit‑card renewal. > "Happened to me months ago, when the credit card failed to renew and it switched back to free, the allow training setting also got activated. Cheeky bastards." – mickelsen
Local‑storage vs server state The setting appears to write a value to localStorage, but the stored value does not affect the UI on new tab loads, suggesting the UI may be optimistic or the backend ignores it. > "I've noticed that toggle sets a local storage entry, but the value of it doesn't appear to matter at all for new tab loads." – antonok
Inconsistent experiences Some users never see the issue; the toggle stays off indefinitely. > "I've disabled it many months ago and it's still disabled today. EU citizen, not sure if that's relevant." – cbg0
Additional opt‑out steps OpenAI also requires a formal privacy request via a separate page ("Do not train on my content"). Turning off the checkbox alone may be insufficient. > "Note: Turning off that checkbox is not enough. You also need to fill out the 'Do not train on my content' request here: https://privacy.openai.com/policies?modal=take-control" – qurren

Community interpretations

  • Skepticism about functionality – Several commenters doubt the checkbox ever worked, given OpenAI’s track record with privacy settings. > "Based on their behavior over the past few years, why would you assume that checkbox even does anything at all?" – sunaurus
  • Possible bugs vs deliberate design – Some argue the flip‑back could be a UI bug, while others suspect intentional back‑door behavior to retain training data. > "Do I cynically believe that these companies keep deliberately and routinely re‑ or un‑checking these checkboxes…?" – jerf
  • Comparison to other services – Users liken the issue to Apple re‑enabling iTunes password sync or Facebook’s shifting privacy defaults, framing it as a broader industry pattern. > "Sounds like the same issue with Apple re‑enabling iTunes sync of passwords." – Tepix

Practical steps for affected users

  1. Verify the toggle – Open the ChatGPT settings page (https://chatgpt.com/#settings/DataControls) and confirm the checkbox state after a page refresh.
  2. Clear local storage – Use browser dev tools to delete the localStorage entry related to the toggle and re‑set it.
  3. Submit a formal privacy request – Fill out the form at https://privacy.openai.com/policies to request that your data not be used for training.
  4. Monitor account activity – After any billing change (e.g., credit‑card renewal, plan downgrade) revisit the setting, as some users report flips coinciding with such events.
  5. Consider alternatives – If the setting cannot be trusted, evaluate competing AI services that provide more transparent data‑use controls.

Open questions

  • What backend API governs the toggle? – Reverse‑engineering could reveal whether the UI merely reflects a client‑side flag or an actual server‑side policy.
  • Does the setting affect other OpenAI products? – Users ask whether disabling training in ChatGPT also disables data collection for Codex, CLI tools, or API usage.
  • Regional differences? – Some EU users report consistent behavior, suggesting possible compliance‑driven implementation variations.

Takeaway

The repeated re‑enabling of OpenAI’s "allow training" checkbox undermines user consent and highlights a lack of transparency in how the company handles data‑privacy preferences. Until OpenAI provides a verifiable, persistent opt‑out mechanism, users should treat the toggle as unreliable and take additional steps—such as filing formal privacy requests and regularly auditing the setting—to protect their data.

Sources

Related