Last week I ended an essay about a quote that lost its caveat with a test: not whether your organization has an AI policy, but whether anyone in it has standing to stop.
I said most governance documents are a PDF nobody has opened since the quarter they were written. I offered that as my own experience, because that’s all it was.
The next day, EY put numbers on it.
Its AI Risk and Governance Survey, released September 15, asked 202 senior AI executives at U.S. public companies with at least $1 billion in annual revenue how they govern AI.
Ninety-eight percent have a formal AI governance policy.
Forty-seven percent say their organization has previously not followed its governance process for an urgent deployment.
Read the second number slowly. The policy exists. Somebody wrote it. Somebody approved it. It was probably presented to a board. And when a deployment was urgent—when the pressure was highest and the scrutiny was lowest—nearly half of these companies went around it.
A policy with an exception for urgency has an exception for the exact situations it was written for.
Everything in marketing is urgent
I have never worked on a launch that wasn’t urgent.
There’s a date, a board meeting, a trade show, a competitor who announced first. Urgency isn’t the exception in a marketing organization. It’s the operating temperature.
So an unwritten rule that says, “We follow the process unless it’s urgent,” doesn’t describe the occasional shortcut. It describes the default.
Grant Thornton’s 2026 AI Impact Survey of 950 senior leaders found that competitor moves are the biggest external pressure driving AI adoption. Many organizations are acting out of fear of falling behind rather than from a clear view of where AI creates value for them.
That’s the same logic I wrote about last week: a senator’s version and a marketing department’s version of the same sentence.
We can’t be careful, because someone else won’t be.
Once that’s the premise, every deployment qualifies as urgent, and governance becomes something you do when you have time.
You never have time.
The stop is the whole point
The most useful finding in the EY survey is one the headline skipped.
Among companies that conducted a formal AI assurance review, 29 percent paused at least a quarter of their AI systems. Twenty-five percent fully stopped at least a quarter of them.
Put plainly: in a quarter of those companies, at least a quarter of what was already running didn’t survive a formal look.
The review gave someone explicit permission to look and explicit authority to say no. That is where the stop finally happened.
That is what governance is.
Not the document. The stop.
This morning’s New York Times newsletter covered AI at its largest scale: the U.N. General Assembly, and whether AI is a bigger risk to humanity than climate change. One line in it applies at every scale: “We don’t have a plan to stop that unknown from happening.”
Most organizations could say the same about much smaller unknowns, like the AI-assisted campaign already sitting in the queue.
Last year I wrote that delegation is not abdication, and that power without oversight isn’t innovation. The new survey data sharpens that point.
Most organizations don’t lack a shutoff on paper. Grant Thornton found that only one in five has a practiced response plan for an AI failure.
The shutoff exists as a sentence. It has never been pulled.
Why this is harder—and easier—for smaller organizations
These surveys are about large companies with compliance staff and review boards. If they skip their own processes under deadline, it’s fair to ask what hope a 40-person family business or regional co-op has.
I think smaller organizations have a different problem, not necessarily a worse one.
There’s no review board to route around. Governance in a small company is whoever is in the room when the decision gets made. Often that’s the founder, who is also the person applying the pressure.
But it also means the person with standing to stop can be named by lunch.
A big company needs a governance framework to find that person. A small company needs a conversation.
And that person matters more than it used to. The AI tools on your team’s desks will agree with the framing you bring them. They may check facts if you ask. They do not reliably originate doubt about the thing you’ve already decided to do.
If the stop is going to come from anywhere, it has to come from a person.
Three questions instead of a policy review
If you want to know whether your organization governs its AI or merely documents it, skip the policy review. Ask three questions.
When was the last time something AI-assisted got stopped before it shipped?
If the answer is never, either everything your team has produced has been right, or nobody is allowed to say otherwise.
Who stopped it, and what did it cost them?
If the person who raised the flag took a hit for delaying the launch, everyone watching learned the lesson. You won’t get a second flag.
What is your urgent-deployment exception?
Every organization has one. Most never wrote it down, which means nobody has decided who gets to invoke it—or what happens after they do.
If you can answer all three with names and dates, you have governance.
If you’re reaching for the PDF, you have a PDF.
I built the Alignment Score to make these questions answerable before a launch instead of after an incident. It’s a six-dimension diagnostic of whether your marketing foundation is strong enough for the AI it’s already using. One of those dimensions, AI Readiness, scores whether you can deploy AI without degrading your marketing, and it caps the score hard if AI is running with no governance or no human checkpoint.
It takes about eight minutes. Take it before your next launch. Then send it to the person on your team who would have to say no: veronicamarkol.com/alignment-score
Sources
- EY, “EY survey finds that autonomous AI implementation outpaces oversight, yielding an AI governance gap,” September 15, 2026. Survey of 202 U.S. senior AI decision-makers at publicly traded companies with at least $1 billion in annual revenue; fielded May 28–June 15, 2026; margin of error ±7 percentage points.
- The New York Times, “The Morning,” September 22, 2026.
- Grant Thornton, “2026 AI Impact Survey Report.” Survey of 950 senior business leaders across 10 industries; fielded February 23–March 18, 2026.