Picture the desk this was aimed at.
You’re in a finance team. An email lands that looks like it came from your chief executive, about a confidential transaction. Then there’s a call. It’s his voice. On some of these, it’s his face as well. He tells you you’re needed on something sensitive, and that lawyers at a large international firm are handling the other side of it. So you ring the lawyers. They talk you through the deal. They tell you to keep it to the people on the call.
Then they give you the account details.
On 5 November the US Attorney’s Office for the Southern District of New York charged three men over exactly that, and put the total at more than $129 million [1].
It’s worth being careful about what that document is. It’s an indictment, the charges are allegations, and the three men are presumed innocent unless it’s proven otherwise [1]. None of that changes what’s useful about it. A prosecutor has written down, step by step, how the money moved. You don’t often get that in public.
What the release actually describes
The method, in the government’s own order [1].
Someone posed as the victim company’s chief executive, using what the release calls “super intelligence” to mimic the executive’s voice, or image, or both, on a call. That’s the government’s term, not mine. Most of the coverage has called it AI.
The fake chief executive told the employee they were needed on a confidential financial transaction, and directed them to contact lawyers at large international law firms who were supposedly also involved. Conspirators then impersonated those lawyers to make the transaction look real. One of the accused is said to have played UK attorneys. The “lawyer” directed the employee to wire the money to an account the group controlled. The funds went to accounts in nine countries, got converted to cryptocurrency to hide where they came from, and were split up [1].
Six victims are named by type in the release. A hospital system in Missouri lost $4,829,580. A hospital system in California lost $2,968,497. A bank in Rhode Island lost $2,684,580. A mutual fund company in Massachusetts lost $4,639,918. A pet food company in Florida lost $5,242,747. A financial technology company in the United Kingdom lost $11,578,474 [1].
Add those six up and you get about $31.9 million. The total charged is more than $129 million [1]. So the named examples are a quarter of it. Reporting of the case says thousands of companies were targeted worldwide, and that the group often went for accounting departments [2].
Look at that list of six again. A hospital, a hospital, a bank, a fund manager, a pet food business, a fintech. There’s no pattern in the industry. The pattern is in the seat. Every one of them had somebody who could move a large sum, and who could be reached.
The script is the part worth noticing
Here’s the detail that I’d put in front of a board before any of the figures.
The release says one of the accused worked from a handwritten script titled “Lawyer Call”, and that the script included language meant to keep the employee from discussing the matter with colleagues [1].
The secrecy wasn’t a side effect. It was written down in advance, because it had to be.
Think about why it had to be. The whole thing collapses the moment the employee turns to the person at the next desk and says “this is odd, has the boss mentioned an acquisition to you?” One sentence to one colleague, and it’s over. The people running it knew that. So they scripted around it.
That tells you where the control actually sits. Not in the email gateway. In whether somebody felt able to ask.
Nothing in this was a technical failure
Go through what didn’t go wrong, because the list is longer than you’d expect.
Nothing was hacked. No malware, no stolen credentials, no unpatched appliance. Multi-factor authentication isn’t relevant, because nobody logged in as anyone else. The law firms were real firms, so a reader checking the name would have found it. The payment went out through the organisation’s own banking, authorised by a person who was entitled to authorise it.
And your people weren’t careless. An employee who’s just been given an instruction by their chief executive’s voice, and told it’s confidential, is doing their job by acting on it. That’s not a lapse in judgement. That’s an organisation chart working the way it’s meant to.
So if you were looking for the technical control that failed here, there isn’t one. Which is uncomfortable, because the technical controls are the ones most of us can evidence.
The controls that would have stopped it are all people controls
There are three, and none of them costs much.
The first is a callback on a channel the requester didn’t choose. Not a reply to the email, and not the number the caller gave you. The number you already had. If the instruction came by voice, you verify by something that isn’t voice.
The second is dual authorisation on payments above a threshold, with the second person free to ask the first why.
The third is the one that actually does the work. A standing rule that an instruction to keep a payment confidential is itself the reason to escalate it, every time, without exception, and with it understood in advance that nobody gets in trouble for applying it to a request that turns out to be genuine.
That third one is a culture control dressed as a procedure. It only works if the person in the seat has heard it recently, believes it, and has used it at least once in a drill. Written in a policy, it does nothing. The script in that indictment was built specifically to defeat a person who had never practised saying no to their chief executive.
Australia has already ranked this, and it’s at the top
If this reads like an American problem, the Commonwealth’s own numbers say otherwise.
On 2 November the Critical Infrastructure Security Centre published the fourth edition of its Critical Infrastructure Annual Risk Review. The risk it ranks first by plausibility, above every purely technical risk on the list, is contracted workforce insider risk [3]. We wrote about those rankings last week, and the argument then was that the government is pointing at your people and your supply chain rather than at your controls. This is what the people half looks like at close range.
The review is specific about the mechanism, too. It says the people doing your hiring now face AI-enabled identity fraud, including deepfake credentials and fabricated references that are hard to verify [3]. That’s the same technique as the fraud above, pointed at a different door. One version calls your finance team pretending to be your chief executive. The other applies for a job.
And it describes AI as having moved from an emerging technology risk in 2023 to a dominant threat multiplier in 2026, embedded across all five of its hazard categories rather than sitting in one, and says traditional security frameworks are no longer sufficient against it [3].
For a bank or an insurer, the same thing arrives wearing a different hat. CPS 230 asks you to manage operational risk and to show that your controls are effective, not just documented [5]. A payment released on a spoofed instruction is an operational risk event with no cyber incident attached to it.
Four per cent
Here’s the figure that should worry you more than the $129 million.
Research published this year by ISACA, the professional body, found that 4 per cent of Australian organisations regularly test how they’d respond to an AI-related incident [4]. Not 4 per cent of small businesses. Four per cent of organisations.
So the technique is in a US indictment with dollar amounts against it. The Australian government has ranked the risk class first for likelihood. And roughly one organisation in twenty-five here has ever rehearsed the response.
That gap isn’t negligence. It’s a sequencing problem. The threat moved faster than the training cycle, and most training cycles are annual.
An annual module can’t produce this reflex
This is the bit where a lot of people-risk programs come unstuck, and it’s not for lack of effort.
Blanket annual awareness training is built to lift a whole organisation off a floor. It does that reasonably well. What it can’t do is prepare a specific person for a specific pressure. The finance controller who can release a seven-figure payment is not facing the same threat as the warehouse supervisor, and giving them the same module in the same month is how you end up with a tested position on paper and an untested one in the seat that matters.
Three things change that, and they’re all operational rather than technical.
Target it by role. Find the seats that can move money, grant access or approve an exception, and give them their own track, pitched at what somebody would actually try on them.
Test the channels an attacker would use. If your testing is email only, you’ve trained people to be suspicious of writing. Nothing in the method above arrives in writing except the first contact. The persuasion happens on a call.
And train at the point of failure, not at the next scheduled course. The lesson has to land while the person still remembers the decision.
Picture who this is, in your business. A financial controller at a mid-sized manufacturer who signs off the payment run on a Thursday. A practice manager at a law firm who moves settlement money most weeks. A finance officer at a water utility with authority up to a threshold nobody has reviewed in three years. A bookkeeper at an accounting practice who handles client funds and reports to a partner who travels. None of those people are the problem. All of them are the pressure point, and in most organisations not one of them has a program built for the seat they’re actually in.
The question for a board is one sentence
If somebody called your finance team this afternoon with the chief executive’s voice, told them it was confidential, and pointed them at a real law firm, who in that team has rehearsed saying “I’ll call you back on the number I’ve already got”, and when did they last do it?
If the honest answer is that they’d have done the right thing anyway, that’s a hope, not a position.
Where BayRisk lands on this
Plainly, so there’s no confusion about what we do.
We don’t design your payments process and we don’t rebuild your finance controls. That’s your CFO’s work, with your bank and your auditor. We won’t tell you otherwise.
What we do is the position around your people. Which seats actually carry the risk, whether the controls around those seats have been tested rather than written, what the testing says over time, and what your board sees. Targeted by role and function instead of spread evenly. Tested against the channels attackers really use, not email alone. And interpreted for the people who have to make a decision about it.
The difference we care about is between a program that’s maintained and a program that’s being run. A maintained one gets attention before a review. A program that’s being run produces evidence as a by-product of ordinary weeks, so when a regulator, an auditor or an insurer asks where your people risk sits, you’re reading from a current position rather than assembling one.
And the risk itself doesn’t move. It belongs to your board and your executive, and it can’t be handed to anyone, us included. What changes is whether the person who owns it is looking at a position or at a completion report.
One last thing, because it’s the easiest thing to get wrong about this story. Nobody in those finance teams did anything stupid. They were put under a pressure that had been designed, scripted and rehearsed by people who do it for a living, and they hadn’t been given the one sentence that breaks it. That’s an opportunity, not an accusation. The sentence is cheap. Not having given it to them is the gap.
If you want to work out which seats in your business carry this and whether anything around them has actually been tested, that’s a short conversation: hello@bayrisk.com.au.
References
1. United States Attorney’s Office, Southern District of New York, “Three Foreign Nationals Arrested And Charged With Scheme To Use Super Intelligence To Defraud Companies Of Over $100 Million”, Press Release 26-317, 5 November 2026 (page updated 6 November 2026). justice.gov
2. PBS NewsHour, “3 charged in $129 million AI scheme targeting thousands of companies”, 6 November 2026. pbs.org
3. Critical Infrastructure Security Centre, Critical Infrastructure Annual Risk Review, Fourth Edition, November 2026, Commonwealth of Australia. cisc.gov.au
4. ISACA, State of Cybersecurity 2026, Australian findings, as reported by a member of the ISACA board in Australian trade press, 30 October 2026.
5. Australian Prudential Regulation Authority, Prudential Standard CPS 230 Operational Risk Management, effective 1 July 2026. apra.gov.au