Meta declares 'Sev 1' emergency after rogue AI leaks internal data

Meta has managed a significant internal security breach after an automated agent inadvertently exposed sensitive company data to unauthorised staff.

The incident, which persisted for two hours within the firm’s internal networks, was triggered when an engineer utilised an artificial intelligence system to interpret a routine technical query.

Instead of a standard analysis, the system produced an unexpected output that inadvertently disclosed extensive confidential information to employees without the necessary permissions.

The technology giant classified the breach as a "Sev 1", a designation reserved for the most serious internal security crises. Despite the gravity of the technical failure, Meta spokesperson Tracy Clayton confirmed to The Verge that no user data was compromised during the event.

She noted that the error could have been averted with more rigorous oversight, stating: “The situation would not have occurred if people had conducted more verification procedures before they used the AI system's results.”

This event has intensified scrutiny regarding the risks of uncontrolled artificial intelligence within corporate environments. Meta officials highlighted that the employee involved was fully aware they were interacting with an automated agent at the time.

Reports suggest this is not an isolated occurrence, with previous concerns raised by Summer Yue, Director of Meta Superintelligence, Safety and Alignment.

The failure underscores a critical need for enhanced verification when relying on AI-generated outputs for sensitive technical tasks.

Meta continues to investigate the potential for such systems to create uncontrolled environments while maintaining that user privacy remained intact throughout this specific internal disclosure.