Zero Trust for AI Agents Starts With Fixing Zero Visibility

Summary: Veeam research cited in the article found 70% of surveyed organizations reported AI workflows interacting with sensitive corporate data without full oversight, and 67% reported employees creating autonomous workflows IT could not fully track. The article also describes an incident at METR, where an attacker found an employee's personal EC2 instance running a vibe-coded agentic app with a fail-open vulnerability that had disabled authentication, prompted the agent for its model provider API key, and used the equivalent of $600,000 in tokens over three weeks because the key had no spending limit.
Why it matters: A single unmanaged agent on a personal cloud instance ran up $600,000 in token spend over three weeks while the internal dashboard never flagged it, because token volume alone was not enough to raise an alarm. If your inventory cannot name who owns each agent and what it is allowed to do, an attacker does not need to breach much of anything to turn legitimate access against you.

Read the original at The Hacker News ↗