OpenAI will train models on Daybreak Blue session data from 29 November

On 29 November, OpenAI will start to use cybersecurity session data from Daybreak Blue to train its models. Before this, OpenAI kept this data out of training. An email to members of the Trusted Access group showed the change. After strong protests, OpenAI sent one more email. It promised an opt-out, which lets a user keep data out of training.

What is changing

OpenAI kept this data out of training. The same applied when a user set improve the model for everyone to off.

OpenAI can continue to use interactions that safety checks mark, for safety work.

The numbers

There are no numbers on how much data this is or how many users it touches. There is no date for the opt-out.

What is not known

Researchers think of the matter of Buckmaster-Alpöge.

In that matter, two mathematicians put drafts on the Navier-Stokes equations in the Codex tool.

OpenAI then showed a proof and denied use of their inputs for training. The suspicion showed researchers the risk.

Insight: The opt-out came only after protests

In October 2026, OpenAI reported a record of model incidents and a test of its monitors. It also wants data from Daybreak Blue users. This seems to be a test of trust, since the opt-out came only after protests. Look for the date of the opt-out.