OpenAI will use work done by vetted security researchers in its Daybreak Blue program to train its models starting November 29, according to a reported email to users that says the policy covers personal accounts even when model training is switched off.
Nov 29 · OpenAI
OpenAI plans to train on security research, including personal accounts with training off
A reported email says work from vetted security researchers in the Daybreak Blue program will be used for model training starting November 29. The change reportedly covers personal accounts regardless of training settings, and an opt-out was promised only after backlash.
Nov 29
Reported start date for training on Daybreak Blue work
Personal
Accounts reportedly covered even when model training is switched off
Flagged
Safety-flagged sessions could still be used for training after opting out
How the reported change unfolds
Step 1
Email to users
Describes training on Daybreak Blue work
→
Step 2
Public backlash
Criticism focuses on personal accounts with training off
→
Step 3
Opt-out promised
Timing and scope not detailed in reporting
→
Step 4
Training begins
Safety-flagged sessions remain eligible
The case for broader training data
High-quality, expert-reviewed security work can improve vulnerability identification.
It may help models handle sensitive technical tasks.
The benefit must still be weighed against consent and data minimization.
The critical reaction
Users who switched training off may have expected their work to stay out of development.
An opt-out that arrives only after public pressure may be seen as inadequate.
How "safety-flagged" material is defined and retained is unclear.
What remains unclear
The exact scope of data collection and user protections await OpenAI's published terms.
No official OpenAI policy document or full email text is available.
The opt-out's timing, scope, and interaction with the safety-flag exception are not detailed.
What Daybreak Blue involves technically is not explained.
Watch for formal policy updates, clarification on flagged sessions, and whether November 29 holds.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article...