You have worked with the same supplier for ten years. The email comes from their usual address, signed the usual way, sitting at the bottom of a real thread about last month’s invoices: “Please note that we have changed banks. You will find our new account details attached; kindly use them starting with this invoice.” The accounts team checks what there is to check. The sender is right, the earlier messages are all there, the amount matches the order. The transfer goes out. Three weeks later the supplier calls to chase the payment, and that money never reached them.
The scene is made up, but anyone who works in an accounts office will recognise it, because it has happened to them or nearly did. This is the bank details change scam, also known as invoice fraud, payment diversion fraud or mandate fraud, and the unsettling part is exactly this: the email was genuine. Nobody had faked an address. Someone had got into the supplier’s mailbox weeks earlier, and had been waiting.
Why the spam filter lets it through
Mail filters are good at spotting what is fake: a sender imitating a domain, a link that leads somewhere it should not, a dangerous attachment. Here nothing is fake. The message leaves a legitimate account, from the right server, inside a real conversation. To any filter it is the usual email from the usual supplier. A good spam filter may stop the bait that started the story; this message gives it nothing to catch. Nor does the text have spelling mistakes or an odd tone, because whoever wrote it had weeks to study how that person really writes.
That is why the defence cannot sit only on incoming mail. It has to sit on the accounts: noticing that someone other than the owner is using a mailbox, before the wrong email goes out.
How attackers get in, even with two-factor authentication
Many people’s first thought is: “We have the code on our phones, this cannot happen to us.” The code helps a great deal, and anyone who has not switched it on should do it tomorrow. But there is a way around it, and Microsoft itself has documented it.
It works like this. An email asks you to sign in: an invoice to download, a warning that your mailbox is nearly full, a document shared by a colleague. The link opens a perfect copy of the Microsoft 365 sign-in page, which sits between the employee and the real service. The employee types the password, receives the code, enters it, and really does land in the mailbox. Meanwhile the go-between has passed everything on to Microsoft and kept the session, the digital pass the service uses to recognise someone who has already signed in. From that point the attacker needs neither the password nor the code: the stolen session opens the door.
Microsoft’s guide to token theft, the technical name for that pass, is clear about it. With the session in hand the attacker does not sign in again, so as far as the system is concerned the code check has already been passed: the victim passed it for them. The same guide lists the signs to look for afterwards: mail forwarding or sorting rules that have been created or changed, and new devices or new authentication methods registered on the account. Back in 2022 Microsoft described a campaign of this kind that targeted more than ten thousand organisations and followed exactly this pattern: into the mailbox, rules to hide certain messages, then payment fraud. In the fastest cases, the post notes, only five minutes passed between the session theft and the fraud attempt.
Timeline of a bank details change scam
The transfer is the last step. The warning signs come earlier, but only someone watching the compromised mailbox can see them.
-
1
The click. Day 1.
A fake sign-in page collects the password and the code, and keeps the session.
Warning sign: sign-in from a network or device never seen before -
2
The hidden rule. Same day.
Replies from certain contacts land in a folder nobody opens, already marked as read.
Warning sign: new mail rule created on the account -
3
The wait. From a few minutes to weeks.
The attacker reads invoices, due dates, names and the way each person writes.
Warning sign: the same session used from a different network -
4
The email. On the right day.
The attacker replies inside the real thread with the new bank details.
To the spam filter it is an ordinary email -
5
The transfer.
The payment goes to the attacker’s account. -
6
The discovery. Weeks later.
The party expecting the money chases a payment that shows as made.
Which side you are on: the scam works both ways
One detail changes the whole picture. In the opening scene the compromised mailbox belongs to the supplier, and the customer pays the price. But it can happen the other way round: the compromised mailbox is yours, and your customers receive from you, from your real address, a request to pay into a new account. You do not get paid, they have already paid, and you are the first one left explaining.
The two directions call for different defences. Over a supplier’s mailbox you have no control at all: there the main defence is a process rule, which you will find below. Over your own mailboxes you can have control: it is a matter of knowing who signs in, from where, and what they do once inside.
How big the problem is, with the numbers we have
The firmest figures come from the United States. According to the 2025 annual report of the FBI’s Internet Crime Complaint Center (IC3), there were 24,768 cases of business email compromise, with reported losses of about $3.05 billion. It is the second largest category by losses after investment fraud, and it includes other email frauds besides changes of bank details. Divide one by the other and the average is over $120,000 per case. These are voluntary reports from US victims, so the real number is higher and does not carry over directly to other countries, but it gives a sense of what each single attack is worth.
Verizon’s Data Breach Investigations Report published in 2026, covering breaches from November 2024 to October 2025, has figures that will feel close to home for many readers. In organisations with fewer than a thousand employees, credentials were among the data compromised in 31% of breaches. In the region covering Europe, the Middle East and Africa, phishing opened the way in 28% of cases, and an unintended action by a person, a click or a mistake, appears in 70%. These are samples gathered by Verizon, not a snapshot of every business: read them as a trend.
In Italy, where we are based, the figure that separates fraud against businesses from other fraud comes from the Postal and Cybersecurity Police, which breaks down the money stolen through online financial crime in the reports of Clusit, the Italian information security association. In 2025, fraud against companies, meaning business email compromise and CEO fraud, where criminals pose as a senior executive, accounted for 5% of the total; in the first half of 2026 it rose to 9%, while making up only 2% of reported cases: few attacks, but heavy ones. The largest share is still online trading scams, at 66%. The police link this fraud to criminals inserting themselves into correspondence between companies, and note that deepfakes are starting to imitate executives’ voices and faces. No official source gives a euro total for this category, and changes of bank details are not counted separately.
What you can do, in order
1. A rule that costs nothing
Any request to change bank details is checked with a phone call to a number you already had, never the one written in the email. That applies to suppliers, and it applies the other way too: tell your customers in writing, once, that you will never change your bank details by email alone. For the direction you have no control over, it is the most practical defence there is, and it needs no tools.
Across the euro area there has been extra help since October 2025: for both standard and instant transfers, the bank checks whether the payee’s name matches the holder of the IBAN and flags any mismatch before you confirm. Companies that pay suppliers in bulk batches should ask their bank whether the check also applies there, because businesses can opt out of it for exactly that kind of payment. A mismatch warning on a long-standing supplier is the moment to stop; a match, however, does not replace the phone call. The final decision stays with whoever pays. How much the check really cuts fraud will show in the 2026 data: Clusit’s March report pointed to this as the first year in which it can be measured.
2. The second factor, done properly
Two-factor authentication remains essential: according to the 2025 edition of the Microsoft Digital Defense Report, more than 97% of identity attacks go through passwords, and phishing-resistant authentication stops more than 99% of them. Not all second factors are equal: physical security keys and passkeys, credentials stored on a phone or computer and unlocked with a fingerprint or a face, hold up even against the fake page, because they do not work on a site other than the real one, while a code sent by text or shown in an app does not. And a session can also be stolen in other ways, for example by malware on the computer. That is why extra settings matter: sessions that expire, sign-ins allowed only from company devices, alerts when something important changes on the account.
3. Someone who notices a compromised account, even at night
The signs in the timeline above are recorded in the mail administration tools, and Microsoft lists them one by one in its guide. The problem is that nobody looks at them: a mail rule created at two in the morning on a Saturday sits there until Monday, and sometimes longer. This is the job of an ITDR service, short for identity threat detection and response: technology gathers and organises the sign-in events, and real analysts check every suspicious move straight away. If the anomaly is harmless, they mark it as resolved. If they cannot rule out an intruder, they block the access and call the company to find out whether it was authorised: better a wasted phone call than a lost payment.
For companies covered by NIS2, the EU cybersecurity directive, there is one more reason. Italy’s implementing decree lists multi-factor authentication and incident handling among the required security measures, and sets tight deadlines for reporting incidents: an early warning within 24 hours of becoming aware of a significant incident, the notification within 72 hours and a final report within a month. Other EU countries have similar timeframes in their own laws. An incident nobody sees cannot be reported within 24 hours. Who falls under NIS2 depends on sector and size, with exceptions both ways: some companies think they are out and are in, and others the reverse. It has to be checked case by case.
Our view
Watching accounts at night and at weekends takes people who are awake at those hours, and skills that only come from doing this work every day. That is why we offer the managed ITDR service of a leading cybersecurity partner, for Microsoft 365 and Google Workspace, with a security operations centre open around the clock: activation, monitoring, response and support are all handled by its specialist team. Many of our customers already use it. And a team that does only this brings the same architecture to many companies, and improves it with what it learns at each one. Experience is shared, data is not: it stays with each company.
No tool, though, replaces the phone call to the number you already had. If you like, we can start with a review: get in touch and we will look together at how your company works today, who makes payments by email, what authentication you use and which rules exist on your mailboxes. It is the same reasoning we apply to cyber security in general, and the one we followed on AI-driven ransomware: attacks change shape, the right questions stay the same.
Frequently asked questions
What is the bank details change scam?
It is a fraud in which the attacker gets into a company’s mailbox, reads it, sometimes for weeks, and then, from the real address and inside a genuine conversation, asks by email for an invoice to be paid into a new account. It is one form of business email compromise, BEC for short. It can hit the payer, when the supplier’s mailbox is the one compromised, or the payee, when it is their own and the requests go out to their customers.
Why is two-factor authentication not enough?
Because with a fake sign-in page acting as a go-between, the attacker can steal the session, the pass the service issues once the password and code have been entered. Microsoft explains this in its guide to token theft: with the session in hand the attacker does not have to sign in again, so the code does not stop them. The second factor remains essential, but it needs to be backed by expiring sessions, sign-ins limited to company devices and monitoring of the accounts.
What is an ITDR service?
ITDR stands for identity threat detection and response: it watches company accounts in the cloud and steps in when someone other than the owner uses them. We explain how the service we offer works on the page about Microsoft 365 and Google Workspace account protection.
What should you do if you have paid into a fake account?
Call your bank immediately and ask for the transfer to be recalled, because the chances of getting the money back shrink by the hour. Then report it to the police, keep the email and any other evidence without deleting anything, warn the supplier or customer involved through a channel other than email, and have the mailboxes checked: as long as the attacker has access, they can try again. With your adviser, also assess whether you have to notify the data protection authority, if the mailbox held personal data, and, for companies covered by NIS2, the national body that receives incident reports, which in Italy is CSIRT Italia, part of the National Cybersecurity Agency.
Further reading
- Microsoft Learn, token theft playbook: how it works, which signs to look for, how to clean up
- Microsoft Security Blog, July 2022: the campaign that went from session theft to payment fraud
- Microsoft Digital Defense Report 2025 (October 2025): identity attacks and phishing-resistant authentication
- FBI Internet Crime Complaint Center, 2025 annual report: complaints and losses from business email compromise (PDF)
- Verizon Data Breach Investigations Report 2026: breaches from November 2024 to October 2025, with the sections on organisations under a thousand employees and on Europe, the Middle East and Africa
- Regulation (EU) 2024/886 on instant payments, which also introduced payee verification for standard transfers
- Clusit Report 2026, March edition: the Postal Police contribution on fraud against businesses in 2025 and payee verification (PDF, in Italian)
- Clusit Report 2026, October update: fraud against businesses in the first half of 2026 according to the Postal Police (PDF, in Italian)
- Italian Legislative Decree 138/2024, article 24: the security measures required under NIS2 (in Italian)
- Italian Legislative Decree 138/2024, article 25: incident reporting deadlines (in Italian)
The average per case is our own calculation based on FBI data. This article refers to obligations under two sets of rules. Italy’s law implementing the EU NIS2 directive applies to entities within its scope that are established in Italy, including those belonging to a group headquartered abroad; companies in other EU countries are covered by their own national implementing laws. The rules on notifying personal data breaches come from the EU’s GDPR: they apply to anyone processing personal data in the European Union, even when the company is based outside it. The information is drawn from the sources listed and is up to date at the time of publication. It is not legal advice: for any specific case, consult a qualified professional.


