One of the things I'm proudest of during my tenure as Mayor of Orting, WA is that we constructed a new, state-of-the-art city hall facility with no public financing or debt. Halfway through that project, we also learned exactly what a modern phishing scam looks like from the inside, and paid five hundred thousand dollars for the lesson.

When I carried on earlier attempts to move this project forward I knew both the challenges we'd face getting the council on board, and the citizens of the community. But as our staff was quite literally working in attics, and annexes; our Police department was outgrowing a facility that was never quite realized as adequate; and our council was sharing time with dance teams, community cooking classes, and the library, the challenges of a community that doubled in size over the course of ten years, after doubling in size the ten years prior to that were clear.

I worked hard with our state partners. My team looked at our current real estate portfolio. And we even worked with ancient insurance policy detectives, to put together a package that divested old property, purchased derelict property on main street, utilized state grants, state appropriations, and just barely…afforded a city hall that would benefit our community for the next hundred years (as the last one had).

Breaking ground was a highlight. It signified community buy in.

The council, the county, the state, and every barrier moving in one direction for our community. Now when I look at the facility and its clock tower, the meeting rooms, the monuments most won't recognize to the historic nature of our community, and the anchor it now is to a rejuvenated main street, I can't help but share pride in that building. But if you were to rewind this story to about halfway through construction, there's a fascinating harbinger of things to come. And if you live in the government space, I bet we could share war stories on what I'm going to describe to you.

I want you to imagine you work in a busy finance department in a city hall somewhere in America. Like many cities, towns, counties, parishes, villages, or jurisdiction – you're probably using Microsoft Outlook to check your email. No doubt you have an excel document to track expenses on various projects or budget items. Perhaps you manually key in data into a database built on software from the late 1980's. (If YOU don't live in municipal government, trust me… if it works it works… even if there's a better way, better ways are NOT the fulcrum on which local government makes decisions.)

In your role, emails come in from various vendors and payees.

... A chip seal vendor bills you via purchase order, $22k.

.... A dance instructor sends in their monthly invoice, $1.5k.

... And perhaps your contracted city hall construction firm sends you an installment request for their monthly regular $500k bill, per the contract.

This $500k bill is normal. It comes from the same person [email protected] on the first Monday of the month at around 9 am. You pay the bill via wire transfer the same day per the instructions in the email and purchase order. The project continues moving forward.

Except this month, you get a weird message from the city manager.

...that's weird… You paid the invoice.

The money has been deducted from the bank account. You know this, the account doesn't sit there the whole time with the whole construction amount, it's part of your monthly financial transactions to ensure everything is lined up for the next month. You know when things are going in and going out.

You KNOW the company has the money. You sent it. It was wired. It went exactly where it was supposed to. So where is it…

We never got a clean answer on exactly how it happened. Best guess, and it's still a guess, somebody got into the contractor's email, or spoofed it well enough that we never told the difference. What we knew for certain was that the money was gone, and gone doesn't come back on its own.

It came back through insurance. Not fast. It took months, filing after filing, proving what we already knew was true, before the reimbursement finally landed.

Does this sound familiar?

When I was on the Association of WA Cities Board, and later the Employee Benefits Trust board I remember getting a brief on the growing prevalence of nuclear phishing scams.

Kind of like torts, there's small claims, there's nuisance, and there's… nuclear. Nuclear is an apt term for holy (expletive) this goes way beyond our ability to simply pay and is going to need a secondary or even tertiary level of insurance (risk management)…

Surprise, your insurance policy has an insurance policy! And that one likely does too.

Well, I can tell you as a small city of about 8,000 people, missing about $500k starts to look pretty nuclear pretty darn quick.

I've told storm stories before. My last article was about AI, storms, and pre-mortems.

How you can anticipate the dangers you already have in your organization, and pivot to a risk anticipation stance rather than an absurd total risk avoidance one.

Those lessons are borne from experiences like this one. While this experience wasn't a natural disaster, if you peel away the actors and the nature of the scenario the main drivers are the same. Something happens that can be anticipated, is happening elsewhere, and yet, there is little you can do to stop it. It's sophisticated, it's persistent, and it's omnipresent.

In reflection that omnipresence is truly the interesting part of this phishing story. The bad actor, whoever they were needed to know the person sending the email. They needed to know when they sent the bill. To whom it was sent. How much to request. The format of transaction. And so on. And this could not have been a one-shot maneuver.

You cannot build that kind of skill on a first try. Moreover, they needed to know the right target; a project bill that is big enough to be worth the effort, but small enough not to be worth the FBI's time to investigate. And to do this they had to do it enough times for it to work at least once, but very likely many times.

And it did. It does. It continues.

But up until recently, this sort of social engineering has been a slow, steady burn of organized crime, or extreme social hacking. On one end, brute force Nigerian email scams evolved to "hey, can you go buy me a gift card" for the modern era.

And on the other end, a guessed email password at a construction company and someone who is personally willing to read through thousands of emails, and quietly build a profile of transactions over the course of a couple months, just to quietly insert themselves into the chain of custody of one payment.

If it wasn't such a pain in the rear, it could even be respectable. Hold onto that idea though, because artificial intelligence just deleted it.

Walk back through the steps and watch what happens when a machine runs each one.

The machine does the research. It reads a city's entire public footprint in seconds, and it ends up knowing more about your vendor relationships than most of your own staff do.

The machine writes the message, and it never fumbles a word. Weird grammar, non-natural English, form statements, the things that used to give these emails away are gone. In fact, the personal anecdotes are even included – the machine in just a few moments searched 45 thousand emails over the course of the last five years and learned to mimic the sender so acutely that even personal jokes between senders and recipients are kept, in context!

The AI writes in clean, confident English, and it matches the tone of a real thread. It sounds so real even the sender can easily mistake it for their own voice. But the machine knows this and minimizes even the potential for accidental discovery. It picks the timing of the attack. The day the construction firm accountant took off. The last email out the prior night wasn't Jim's sign off, it was a scheduled request set up by the AI to send at the perfect time, unsupervised.

A perfect mimic that cost the thief pennies, if that. This can all be run on local machines, and open-source models.

Put those facts together and the scale of what we're discussing goes… well, what's beyond nuclear? In simple terms, the math of the crime inverts.

The con that took a skilled human days or weeks to build against one city now runs against every city in the country at the same time, for the price of the electricity.

The criminal no longer must choose you, because choosing is free.

There is no craftsman to run out of hours. There is only a program, running the same patient, personalized con against ten thousand finance departments at once, and only needing a handful of them to say yes.

Targeting used to be the scammer's bottleneck. That bottleneck is gone.

Here is the part that should keep you up at night, because you built it yourself, I built it. The law requires it, generally, for good reasons.

The man who took our money had to dig for what he knew. He had to work to learn our contractor, our project, our dollar figures. An AI system does not dig. It reads. And you already publish everything it needs.

Think about what lives on your website right now. Your contracts, with the vendor names and the amounts. Your council minutes, which narrate the project, quote the company, and print the dollar figure into the permanent record. Your budget. Your staff directory, with names, titles, and email addresses arranged in a tidy org chart. Your meeting videos. The very transparency that makes your government accountable to the people is now the training manual for the machine that intends to rob them.

You cannot un-publish any of it, and you should not try. Open records are the whole point of open government. You can try to fight that battle if you want. You will lose. I argue, like the winter storm, this is a storm you know about and need to re-orient your assumptions for.

This new scammer is faster than the last one. They are more subversive. And they are targeting you by name, and your peers as well. They have every word you've ever said available to pull from and can even mimic your voice. Your protocols are already broken, you just don't realize it yet.

For years the standard defense was simple. If an email asks you to move money, pick up the phone and confirm it with a human voice. That advice was good. It is now dissolving in front of us, and I want you to understand why.

Your finance director spoke at a council meeting that is sitting on a public video platform right now. Thirty seconds of that audio is enough for a machine to clone her voice well enough to fool the people who work with her every day. The voice on the phone is no longer proof of anything.

This is not theoretical. In 2024, a finance employee at a global engineering firm wired about twenty-five million dollars after joining a video call with colleagues he recognized. Every person on that call, including the chief financial officer, was an AI deepfake built from public footage. He was not reckless. He saw familiar faces, heard familiar voices, and did what they asked. The tools that fooled him are cheaper and better today than they were then, and they keep improving.

The phone call you thought you could trust. The video meeting you assumed was safe. The voice you have known for years. Treat all of it as evidence now, not proof.

You might be tempted to file my story under bad luck. DON'T!

The pattern is documented, it is enormous, and it is accelerating.

The FBI has a name for this category of crime. It calls it business email compromise, and it is not a rounding error in anyone's budget. These scams have cost victims more than fifty-five billion dollars since 2013, and they took two point eight billion dollars in 2024 alone. That is not a fringe threat. The FBI ranks it among the costliest online crimes it tracks.

And as I learned to my misfortune, local governments like mine were and continue to be squarely in the crosshairs.

Different states. Different years. Identical con.

The saying misery loves company does not seem apt.

Mine was one of the early, hand-built versions, back when the crime still took a skilled human.

What is coming for you is the automated one. It costs the criminal almost nothing to run, and it is pointed at everyone at once. What are we to do?

Local government has already lived through a sort of parallel to this. Though somewhat more ham handed.

You may recall, ransomware once swept through cities and counties, locking up their systems, and forcing impossible choices in front of the whole community. You still likely read about this every once in a while and say, "that's why I never click on a link I don't know." Good job. Thumbs up.

What was the solution reflex?

That reflex was wrong then, and it is wrong now.

Ransomware was never an IT problem. It was a governance problem wearing an IT costume.

The leaders who came through it well were the ones who treated it as their own risk to own, not a box for the technology department to check.

They ran the drills. They made the hard calls about backups and payments before the crisis, not during it.

The scam I am describing works the same way.

When the money leaves the building, "the software let it happen" is not a defense you will offer your council, and it is not a defense you would accept from anyone who works for you.

You do not get to delegate this to IT, because IT cannot stop your own finance clerk from following a convincing instruction from a trusted vendor. This is a governance failure waiting to happen, which makes it yours.

I have written before that the executive's real job is not to be the visionary. It is to be the Chief Risk Officer. This is exactly the kind of risk that title exists to carry.

And spare yourself the comfort of "we are too small to be a target." That single sentence is what wrecked small towns during the ransomware wave, it put my own community in jeopardy, and it will do it again to you.

In the long ago times of say…three years ago, a human con artist had to decide you were worth the hours. Small towns often were not, and that math protected you.

AI makes no such decision, because there is no effort to weigh. The program does not care whether you are Los Angeles or a town of four thousand. Small is a target now. Everyone is.

So do the thing I keep coming back to in this article and in my newsletter and in my advisory. Run the pre-mortem.

Do not wait for the loss and then assemble a post-mortem to explain it.

By the time you hold a post-mortem, the money is already gone and the damage is already done. Get ahead of it instead. Put your finance team, your clerk, and your top staff in a room, and open with a single sentence. "Last month, we wired five hundred thousand dollars to a criminal pretending to be a vendor, and nobody caught it. Tell me how."

Then make them work backward, out loud, until the failure has nowhere left to hide.

How did the request get in? Who received it, and what made it look normal? What made the change of payment details believable? Who had the authority to approve it alone, and why was one person enough? What check did we think was happening, that never happened? At what exact step could a single "wait, let me verify that" have stopped the whole thing?

Push them past soft language. If someone says "we would probably catch that," make them show you the specific control that would do the catching. If it does not exist, you have found your gap.

Name each failure plainly, write it on the wall, and only then start writing the rules that would have closed it. A pre-mortem is not a meeting about fear. It is a meeting about exactly where you are soft, held while you still have time to harden.

Some of the rules that come out of that room are not complicated. They are cheap, they are boring, and they work. These are the ones I would insist on.

We paid five hundred thousand dollars in Orting to learn something that, in my day (four years ago), a criminal had to earn through real skill and real hours.

The next city will not get the courtesy of a human on the other end. The next con will be generated, not crafted. It will be cheap, it will be flawless, and it will be aimed at every city in the country at once. The scam that used to require a master forger now requires a ChatGPT subscription, at most.

You cannot stop the attempts from coming, any more than I could stop the snow from falling on Orting.

But that was never the job. The snow was going to fall whether we were ready or not. The failure I could actually own was the resident we did not reach in time, and the failure you can own is the payment that should never have cleared.

So, decide it now, on a quiet Tuesday, in a conference room, with coffee and a whiteboard and no crisis in the room. Decide exactly how your city will refuse to pay. Ask the question before the email ever arrives. How will we have failed? Name it, write the rule that stops it, and close the door the con walks through.

Do that, and the worst day never makes the news.

Which, in this work, is the whole point.

"The AI Policy Every Local Gov Needs Before It's a Headline" is a free hour on September 3. It is the same pre-mortem posture applied to the tools your staff are already using, and you leave with a policy skeleton you can hand to your council or your attorney. If you lead a city or a county, or you brief someone who does, come sit in.

If you want to go further than an hour, the four week course for local government leaders, Responsible AI for Local Government, opens September 15.

Both are at maven.com/joshuapenner.