OpenAI will train its AI on cybersecurity experts’ chats, even if they refuse

OpenAI has announced a controversial change to its data policy for members of the Daybreak Blue trust program, effective November 29. Unlike previous arrangements, content generated during these cybersecurity sessions on personal accounts will be used to train its models, even if the user has explicitly disabled the option to improve the model for all. This decision breaks the prior guarantee that such interactions remained isolated from training, sparking an immediate reaction of distrust among security specialists. This issue is critical for open data and algorithmic transparency, as it introduces significant opacity regarding which data feed artificial intelligence systems. Daybreak Blue sessions may contain highly sensitive information, such as unreported vulnerabilities and specialized attack techniques. By allowing these proprietary findings to become part of the training dataset, OpenAI raises serious risks to confidentiality and intellectual property, calling into question the ethical standards of using data without explicit and detailed consent in this specific context. Finally, although OpenAI promised to offer a specific opt-out option and clarified that the change does not affect enterprise customers or secure processing modes, the practical implementation remains uncertain. The lack of clarity regarding the technical details of how the opt-out will function, along with the exact start date, has exacerbated concerns within the community. This case underscores the urgency of robust and transparent mechanisms for handling sensitive data, demonstrating that ambiguity in privacy policies can erode trust in technology platforms and negatively impact open collaboration in high-security fields.

Source: infobae.com
Published on 2026-11-01

OpenAI will train its AI on cybersecurity experts’ chats, even if they refuse

Related news