OpenAI notified Trusted Access program members via email that it will begin using Daybreak Blue cybersecurity session data to train its models starting November 29. The data was previously excluded from model training, even for users who had disabled data-sharing settings.

Following swift backlash from the community, OpenAI sent a follow-up email promising to offer a separate opt-out mechanism for affected users. However, interactions flagged during safety reviews will remain accessible for internal safety evaluation purposes.

The policy shift has raised concerns among researchers about potential intellectual property and data privacy issues. The reaction reflects lingering industry apprehension stemming from past incidents, such as when researchers questioned whether code inputs to OpenAI tools were leveraged without explicit permission.

Why it matters

  • Enterprise and security teams must re-evaluate opt-out policies to prevent proprietary data exposure in model training.

  • Developers using sensitive cybersecurity data must monitor platform data retention rules closely.

  • Policy changes around previously restricted datasets create trust friction between AI providers and technical communities.

Source: the-decoder.com