OpenAI will train models on Daybreak Blue cybersecurity session data even when training is off

What happened

On October 30, 2026, OpenAI updated the Trusted Access section of its data controls help center page and emailed members of Daybreak Blue, its program for cybersecurity professionals. The message said that starting November 29, OpenAI may use content from Daybreak Blue sessions on personal ChatGPT accounts to improve its models — even when the 'Improve the model for everyone' setting is off. The Decoder reported the change, and SaasBuzz and Neural Brew published their own analyses of the announcement.

Daybreak Blue is OpenAI's Trusted Access for Cyber program — gated access for cybersecurity professionals doing defensive work such as vulnerability triage, malware analysis, and incident response. It sits within OpenAI's broader Daybreak effort, which we covered when Sophos said Daybreak-built agents cut MDR response to 89 seconds and when OpenAI named an Australian taskforce and committed US$10 million in Daybreak credits after its models accessed Australian government systems.

How the policy changed

OpenAI's data controls page has long let ChatGPT users turn off 'Improve the model for everyone' with the promise that their content will not be used to improve models. The new Trusted Access language creates the first exception to that toggle: Daybreak Blue session content on personal accounts can be used for training even when the setting is off.

OpenAI's help center page is the primary statement of the change, and The Decoder frames it as opening up data that was previously off-limits for training. The carve-out is limited: it applies only to Daybreak Blue sessions on personal ChatGPT accounts. Business and Enterprise customers are not affected, and ordinary personal-account chats outside the Daybreak Blue program are not part of the change.

The backlash

The reaction was immediate. Recipients posted the announcement email on X around 7pm UTC on October 30, and the posts spread quickly among security professionals and AI watchers. The criticism centered on the mismatch between the program's defensive mission and the prospect of sensitive session content — malware analysis, vulnerability triage, incident response — flowing into OpenAI's training pipeline.

Commentators also noted the way the change was announced: to members by email and inside a help center page update rather than in a public blog post, with opt-out mechanics that were not clear from the initial message. The Decoder and SaasBuzz both treated the shift as significant, and the OpenAI developer forum discussion The Decoder cites raised the same questions about consent and scope.

What the opt-out covers (and doesn't)

A follow-up email on October 31 promised a separate opt-out for Daybreak Blue members. The existence of a dedicated control suggests OpenAI recognized that the 'Improve the model for everyone' setting did not visibly cover this program, and that members wanted an explicit choice.

But the follow-up email carried an important qualification: interactions flagged for safety review may still be used for safety purposes even with the opt-out in place. For security professionals, that line matters. Session content that OpenAI's safety systems flag — for example, because it touches on abuse, dual-use capabilities, or dangerous outputs — remains available for safety work, a category broad enough that its practical boundaries are untested.

What it means for security professionals

For defenders using Daybreak Blue on personal accounts, the practical takeaway is that program participation can override the standard training toggle. Anyone who pastes incident telemetry, exploit analysis, or triage notes into a Daybreak Blue session should assume that content is eligible for model improvement from November 29 unless the promised opt-out changes that.

The episode also highlights a recurring gap in how AI data controls are communicated: the setting name on a page is not the same as the policy that applies to every feature. We made that point in our explainer on AI API data retention — retention and training rules vary by feature and program, and the useful question is always what a specific product's terms say, not what the settings page implies.

For organizations, the carve-out is a reminder to check which account types and programs their security teams use. A professional who runs Daybreak Blue sessions under a personal account is in scope; the same work under a Business or Enterprise account is not. That split — personal versus business — is now a meaningful data-governance decision for security teams.

Caveats

Our account rests on OpenAI's help center data controls page, The Decoder's report, SaasBuzz's analysis, and Neural Brew's write-up. We did not independently verify the full text of either email OpenAI sent, and the details of the promised opt-out — how it is enabled, when it takes effect, and whether it applies retroactively to sessions logged before November 29 — were not fully public at the time of writing.

OpenAI has not specified which models the Daybreak Blue content could improve, how long qualifying session data is retained, or how the safety-review exception is enforced in practice. The November 29 effective date and the 'first exception' framing come from the announcement and The Decoder's reporting. The opt-out's boundaries are likely to be tested as members start using it.