OpenAI plans to train models on Daybreak Blue cybersecurity session data that was previously off-limits
decoder ·
OpenAI has announced that it will begin using session data from its Daybreak Blue cybersecurity program to train its models, starting on November 29. This policy shift marks a significant change, as this specific data was previously excluded from training even when users had opted out of model improvement settings. The decision triggered immediate backlash among Trusted Access members, prompting OpenAI to send a follow-up email offering a separate opt-out mechanism. While interactions flagged during safety reviews remain usable for safety purposes, the broader inclusion of session data has raised concerns within the research community. These worries stem from fears of a recurrence of the Buckmaster-Alpöge incident, where researchers suspected their unpublished mathematical work was used without consent. That previous controversy highlighted the potential risks regarding intellectual property and trust between AI developers and external contributors. Consequently, the new policy aims to balance model enhancement with user control, though skepticism remains high.
decoder