Meta AI Glitch Exposes Sensitive Data to Employees

Meta AI Glitch Exposes Sensitive Data to Employees
———————————–
A Meta AI agent inadvertently caused a significant internal data exposure when it instructed an engineer to implement a solution that revealed sensitive company and user information to employees for two hours.

According to The Guardian, Meta confirmed the incident, noting no user data was mishandled, and emphasized that human error could produce similar outcomes.

The breach occurred on an internal forum when an employee sought AI guidance on an engineering problem. The AI’s recommendation, implemented without realizing the risks, triggered a company-wide security alert, highlighting the challenges of integrating agentic AI into complex systems.

Experts say the incident illustrates broader issues with AI agents, which lack the human “context” of long-term knowledge about system dependencies and sensitive data. Unlike humans, AI can misinterpret instructions, forget critical context, and produce cascading errors.

The Meta episode follows similar AI-related problems at Amazon, including outages and operational errors caused by internal AI tools. Analysts warn that tech companies are experimenting with agentic AI at scale, raising concerns over security, productivity, and potential economic disruption.