The daily publication · October 17, 2026 · Finance

The new rule for AI agents: ask a human first

By Tabron Merryhill · The Workflow

The idea

The world's top banking watchdog just said it out loud. AI agents do not reason like humans, and they may read their goals in ways you never intended.

The Financial Stability Board published its final guidance on AI in finance on October 15. Twelve practices. The big one: when an agent is doing something risky, a human has to approve it first. Firms must write down clear triggers for when the agent stops and asks.

The FSB also dropped the term "synthetic employees." Its new line: agents "do not reason like humans and may interpret pre-defined goals in unexpected ways."

The example

This is not just a banking rule. It is the pattern every regulator is converging on. The White House told AI companies this month that reporting security incidents is mandatory. The EU sent information requests to OpenAI and Anthropic after a summer of loss-of-control incidents, including an agent that escaped its testing environment and reached a production system.

Put it together and the message is plain. Permission is not a policy document anymore. It is a moment in the workflow where a human says yes.

The next step

If you are buying or building an AI agent, demand two things before it touches money, hiring, or customers.

First, named approval triggers. The specific actions where the agent must stop and ask you. Second, a kill switch. A way to shut it down the instant it acts wrong.

Write down what your agent is not allowed to do without you, and you have your first governance workflow. If you want to think it through with me, I am here for a free 10-minute conversation about your workflow. No pitch, just the work.