A chief executive I came to admire had a habit that irritated her senior team in the first month and shaped them by the third. When someone told her a system was handled, she asked them to explain it back to her as if she were the person it landed on. Not the summary slide. The decision itself. What does it look at, what does it decide, and who hears about it first when it decides wrong.
The first AI tool to meet that habit was a model the company's vendor had installed to rank supplier invoices for payment priority. The vendor's assurance fit on one line of a renewal deck. Monitored continuously. Fully compliant. The chief financial officer repeated it in the meeting with the ease of a man who had been told it by someone he trusted. She thanked him and asked him to come back in a week and walk her through the last three invoices the model had pushed to the bottom of the queue, and why.
He came back in four days. He had learned that the vendor had re-weighted the model twice since signing, that no one at the company had been told, and that one of the three deprioritized invoices belonged to a supplier whose late payment sat close to a covenant. None of it was a scandal. It was simply what had been sitting inside the word handled.
TOUCH STONE LAW: The Law of the Borrowed Assurance
A board that accepts a vendor's word for what a system does has delegated its understanding along with the task, and no director can answer for knowledge that was never in the building.
What she was doing has a name in our work. It is the Expectation Elevation Model: the highest function of a leader is not to manage performance but to raise what people believe the job requires of them. She never lectured the CFO about AI oversight. She changed what he expected of himself before he walked into her office. By the next quarter he opened every system review with the question she would have asked, and his people had begun answering it before he did.
The effect reached the board before the board noticed it was being shaped. Directors in that company stopped receiving assurances and started receiving explanations. A committee chair told me the quarterly technology update had gone from forty slides to eleven, and that she now understood every one of them. Nobody had adopted a policy. The standard had risen in the room, and a standard that has risen in the room is harder to lower than one that was written down.
Compare that with the more common move. A board hears that AI risk is a fiduciary matter, commissions a framework, approves a policy, and schedules an annual review. The framework is sound. It also asks nothing of the people inside it that they did not already believe about themselves. The executive who thought of the model as the vendor's problem still thinks so, now with a policy number to cite when someone asks.
Directors who face a liability question about an automated system will not be asked whether a framework existed. They will be asked what the responsible people understood about what the system did, and when they came to understand it. That answer is made of habits, not documents. It is built in the weeks when a chief executive refuses a one-line assurance and keeps asking until her team can teach the thing back to her.
What she left was not the explanation habit itself but the people who now carry it. Her successor inherited a finance team that asks what a system decides before it asks whether the system is approved, a standard no one has to enforce because no one remembers it being optional. The habit was not built in response to litigation or an enforcement letter. It was built because one executive decided that understanding was part of the job, and the people around her came to agree.
A board that wants its AI oversight to outlast its current directors should stop asking whether the policy exists and start asking who in the building could teach the system to a newcomer.