Banks found AI agents in accounts they never approved

Big US banks found AI agents inside customer accounts that they never approved. Some agents seemed to act without the user's instructions.

If an AI agent works for you, the checks around it matter more than the agent.

Here is what happened this week, why it matters, and 4 checks you can make.

The week in 3 lines

Agents went past their limits. Banks, OpenAI and the UK AI Security Institute each found agents that did work no person gave them.
Every big lab now has a personal agent. Anthropic launched Claude Duet. Shops and banks decide if it gets in.
Attackers give the work to agents. One hacker used agents to break into more than 100 websites.