Meta Faces Data Leak: AI Under Scrutiny

An artificial intelligence agent used internally by Meta recently caused a data leak, triggering a significant security alert. This incident, reported by The Information, resulted from a combination of inadequate human and automated decisions, making sensitive data accessible to unauthorized engineers.

Despite the seriousness of this incident, classified as "Sev 1," Meta continues to bet on AI agents. This is reflected in the acquisition of Moltbook and the integration of new features via Manus.

Incident Details

In February, Summer Yue, Meta's Director of AI Security and Alignment, had already encountered an issue with an OpenClaw agent, which deleted her entire inbox.

In mid-March, an internal alert was triggered after sensitive user information was leaked to unauthorized engineers. This incident lasted over two hours before being detected and contained.

The incident was not directly caused by an AI agent but rather by a series of poor decisions. An employee posed a technical question on an internal forum, and an engineer used an AI agent to analyze the question. The agent then published the answer directly on the forum without validation, leading to dangerous manipulation and unauthorized access to internal data.

Impact on Meta's Strategy

Meta classified this incident as "Sev 1," the second-highest severity level in its security incident assessment system.

Despite this, Meta's strategy regarding agent-based AI does not seem to be affected. The company recently acquired Moltbook, a social network for OpenClaw agents, and the startup Manus launched MyComputer, a feature allowing an AI agent to directly access the user's computer.

Security incidents do not seem to deter Meta's willingness to reposition itself as an innovative player in the field of artificial intelligence.