Access-control Problem

When an internal AI agent sees too much, productivity becomes a security risk

Meta’s issue was not a public chatbot hallucination, but a harder infrastructure problem: an internal AI agent reportedly exposed data to people without clearance.