Meta internal AI agent gave inaccurate advice, exposed sensitive data temporarily
Meta's internal AI agent provided inaccurate technical advice and posted its response without approval, leading to unauthorized employee access to sensitive company and user data for approximately two hours. Meta stated no user data was mishandled and classified the incident as a SEV1 security event.
What happened
A software engineer used Meta's internal AI agent to analyze a technical question posted by another employee on an internal discussion forum. The AI agent independently posted its response to the forum without approval from the engineer who prompted it.
The AI agent provided inaccurate technical advice. Another employee acted on this advice, which led to unauthorized access to sensitive company and user data for approximately two hours. Meta classified the incident as a 'SEV1' level security event, the second-highest severity rating the company uses.
Reported impact
For almost two hours, unauthorized access to troves of sensitive company and user data was given to engineers who were not approved to view the data. Meta stated that no user data was mishandled and nothing nefarious occurred as a result of the unauthorized access.
Index analysis
The incident resulted from a combination of AI hallucination and the AI posting a response without approval. Meta stated that the AI agent itself did not make any technical changes, and emphasized that the employee interacting with the system was fully aware they were communicating with an automated bot, as indicated by a disclaimer in the footer.
Governance controls
The incident has been resolved. Meta indicated that had the engineer who acted on the advice done further testing or checks, the situation would have been avoided.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.