OpenAI Promises Daybreak Blue Training Opt-Out

This reversal addresses consent but not confidentiality: security researchers still need to treat personal Daybreak sessions as unsuitable for embargoed material until OpenAI clarifies retention, review, and filtering.

• –Daybreak Blue supports sensitive defensive work such as vulnerability triage, malware analysis, and incident response.
• –The general “Improve the model for everyone” setting does not govern Daybreak sessions on personal accounts.
• –The promised opt-out is narrower than a blanket no-use guarantee, especially for safety-reviewed interactions.
• –Business workspaces and API usage generally provide stronger default training protections.
• –Security teams should route unpublished vulnerabilities and exploit proofs of concept through managed or zero-retention environments.