OpenAI Codex Training Toggle Silently Re-Enabled for Users

Developers using OpenAI Codex report that the platform silently re-enabled data training settings on November 3, 2026, despite users having previously opted out. A Hacker News thread garnered 345 points and 103 comments as enterprise teams discovered the 'Improve the model for everyone' toggle had flipped back on without notification. OpenAI has not publicly addressed the incident or confirmed whether the change is intentional or a bug.

What Developers Are Reporting

The top-ranked Hacker News post, submitted by user052919, describes how multiple team members at a single company noticed the setting had been re-enabled. "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," the post reads.

Commenter chriskanan confirmed 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." Another developer who built internal monitoring tools noted the severity: "A pretty serious issue since we tell our customers that none of their data is being trained on." The toggle in question lives at chatgpt.com/settings/data-controls and controls whether user interactions contribute to model improvement.

Enterprise IP Security Implications

The incident lands as enterprises tighten policies on proprietary code exposure to coding assistants. Companies relying on explicit opt-out guarantees now face uncertainty about whether their source code and technical discussions have entered training pipelines. One Hacker News commenter, monster_truck, offered blunt skepticism: "If your company actually cared about this they would be running their own inference instead of leaving fate to a couple of checkboxes."

The report follows a separate disclosure that Postman Desktop installs plugins into Codex and Claude Code without user prompting, adding to concerns about silent behavior changes in developer tools. Commenter kevcampb reported a similar issue with Claude last week, where feedback mechanisms appeared to trigger automatically despite organization-level settings disabling them.

OpenAI Response and Transparency Questions

The Hacker News thread shows no official OpenAI response as of November 3, 2026. The company has not issued a statement confirming whether the toggle reset is a bug, a policy change, or an isolated incident affecting a subset of accounts. The silence contrasts with OpenAI's public commitments to enterprise data controls and transparency around training data usage.

For organizations using Codex under the assumption that proprietary code remains excluded from training corpora, the lack of acknowledgment creates compliance and audit challenges. Legal teams at firms with strict IP safeguards may need to verify whether any data entered during the window when the toggle was active has been retained or used.

Availability and Current Status

Officially launched on November 3, 2026, this incident affects users who had previously disabled the data training toggle in their Codex settings. The toggle itself remains accessible at chatgpt.com/settings/data-controls, though users report needing to manually re-verify the opt-out status. No timeline has been provided for a formal explanation or remediation from OpenAI.

What This Means

The silent re-enabling of data training settings in Codex underscores the fragility of user-controlled privacy mechanisms in AI coding assistants. Enterprise teams relying on explicit opt-outs must now implement additional monitoring and audit trails to ensure settings persist across platform updates. Until OpenAI clarifies the root cause and publishes safeguards against future resets, organizations may reassess their risk tolerance for cloud-hosted coding assistance. The incident also highlights the broader challenge of maintaining data sovereignty when proprietary workflows depend on third-party AI services.