An email to Trusted Access members said the data had previously been excluded from training, even when users had turned off “improve the model for everyone.” After backlash, OpenAI sent a second email promising a separate opt-out.
Researchers are concerned about a repeat of an earlier dispute involving mathematicians’ Navier-Stokes drafts uploaded to Codex. OpenAI denied using those inputs for training; the report presents the case as context for concerns about the Daybreak Blue policy.
Why it matters
The change raises questions about consent and trust when research-session data may be used for model training.
Something to try
Hypothetically, if you use such a program, locate the new opt-out and the safety-review exception, then check what data may be used and for which purposes.
Start the conversation.