By @ACSnewsfeed
Publication Date: 2026-03-25 02:39:00
“AI agents have neither the maturity nor the context to assess the impact of their actions,” one expert said. Photo: Shutterstock
Software engineers need to maintain a healthy skepticism of the advice of AI agents, Meta said after a major security incident caused when an AI agent gave incorrect technical advice – and a human engineer followed it.
The incident occurred when an internal AI agent at Meta discovered a technical question that had been posted by an employee on an internal forum and then posted an answer without waiting for the employee’s approval.
When another employee read the post and followed the “inaccurate information” it contained, the agent inadvertently gave engineers access to a large amount of user data and company information that was not authorized to view.
The incident – which lasted almost two hours before it was discovered – was classified as “SEV1”, the…
