Short answer. If a company mailbox has been hacked, seven actions within the hour: change the password from a clean computer, close every unknown session and device, delete the rules and forwards that were added, turn on two-step verification and check the recovery number, warn the bank if a payment has gone out, warn customers and suppliers by phone, keep the log and the messages. Then two rules close the door: a second proof at sign-in for every account, tied to the site for management and accounts, and no change of bank details on an email alone: a call back on the number already known, then a second approval.
This guide is for the manager, finance director or accountant of a company that pays its suppliers by bank transfer, local or foreign, for anyone who has just discovered a suspicious message sent in their name, and for the founder based in France whose Algerian company receives and sends its invoices by email. The mechanisms described come from public sources: business email compromise, mailbox rules used to hide messages, the relay-page phishing described by Microsoft in 2022, and the “account hacking” sheet from the French government site Cybermalveillance.gouv.fr. The numbers in the figures are a constructed example: an accounting mailbox opened by a third party for thirty-one days, and an invoice for 3,240,000 dinars. This guide gives no legal rules: filing a complaint, recourse with the bank and what the company owes third parties are matters for your legal adviser.
Six real situations, and what to do
Each row is a sentence said on the day you understand, and the answer is an action or a rule, not one more piece of software.
We received new bank details from the supplier by email: this is the most common form of what is called business email compromise, a fraud that uses email to deceive an organisation. The FBI's complaint centre, as cited by Wikipedia, recorded 24,768 complaints of this kind for 2025 and a little over three billion dollars in reported losses; these are American figures, given for the order of magnitude. Customers received odd messages from us: the mailbox is then used to deceive others, by sending them your forged bank details; it is your receivables that go elsewhere, and customer credit follow-up is what reveals it. The mailbox gets fewer messages than before: a rule moves some messages, marks them as read or forwards them; the technique is documented, and it is the first place to look.
We changed the password, so it is sorted: not while the session opened by the intruder and the rule they created are still there. Yet he had two-step verification: a forged sign-in page can sit between the person and the real site, relay the password and the code, and keep the session. In 2022 Microsoft described a campaign of this kind that had targeted more than ten thousand organisations since September 2021, followed by payment fraud. Hence the value of a second proof tied to the site itself, described below. The supplier is chasing an invoice already paid: that is often how everything is discovered, weeks later; the bank is warned within the hour, because that is when a transfer still has a chance of being stopped, with no guarantee.
The email thread: five messages, only one is forged
The figure shows the conversation as it appears in the accounting mailbox, and the two things that cannot be seen there.
What can be seen. An invoice, an acknowledgement, then a message announcing a change of bank, and a transfer. What cannot. First, the reminder the real supplier sent five days ago: a rule created in the mailbox marks their messages as read and files them in a folder nobody opens. Second, the address of the fourth message: fournisseur-exenple instead of fournisseur-exemple, a domain registered for the occasion, one letter away. The message replies inside the thread and reuses the signature and the invoice number, because whoever writes it has been reading the mailbox for thirty-one days.
Four tells. A look-alike domain, checked by reading the full address and not the displayed name. A change of bank details announced by email. Urgency, and the request not to phone: the phone call is what makes the fraud fail. A reply that is perfectly on point, which reassures when it should alarm. What domain signing does not do. SPF, DKIM and DMARC, described in the guide to professional email, stop anyone imitating your exact domain; they can do nothing against a look-alike domain, which is another domain, in good standing with itself. They remain necessary: without them, the fraudster does not even need to register a domain.
«The forged message looks nothing like spam: whoever writes it has been reading the mailbox for a month.»
What the mailbox log was already saying
The figure sets side by side the sign-in log, which the mail service had been keeping all along, and the seven actions of the first hour.
The log. Thirty-one days before the discovery, the accountant clicks a forged shared-document link and types her password. Eleven minutes later, a sign-in succeeds from an unknown device, abroad, with no second proof; six minutes after that, the rule is created. Then reads, at night: two hundred and twelve messages in one go, and the invoice thread on the evening the forged message went out. Four lines out of seven should have raised an alert; none had been set. Your mail service keeps a log of this kind; the question is who reads it.
The first hour. The seven actions follow the “account hacking” sheet from Cybermalveillance.gouv.fr, which recommends changing the password immediately, checking that there is no redirection or filtering rule, disconnecting any unknown device or session, turning on two-factor authentication, checking the recovery number and address, and warning your contacts and your bank. The order matters: the password first, from a computer you are sure of; then the sessions, because, depending on the service, an open session survives a password change; then the rules and forwards, which survive everything. Nothing else is deleted: the log and the messages are what will make it possible to say what was read. If the same password was used elsewhere, it is changed everywhere. And if other computers are affected, move to the first-hour sheet for an attack.
Changing a supplier's bank details, in six steps
The second step is accented: ten minutes on the phone, on the number you already had.
Receive the request without replying to it, whatever the channel. Call back the supplier on the number recorded in their file before the request, never the one the message gives, and speak to the person you know. Get it confirmed in writing, through the usual channel, on a letterhead document. In the example, that single call would have taken ten minutes; it would have avoided a transfer of 3,240,000 dinars.
Approve in pairs: the person who enters the new bank details in the software is not the one who approves them, and a change to a supplier record is logged, as in the guide to roles and the audit log. Wait one working day between the change and the first payment, time for the supplier to react if they asked for nothing. Log who asked, who called back, who approved. The rule applies to everyone, including when the message seems to come from the manager and asks for an urgent, discreet transfer: that is the other form of the same fraud. And it applies in the other direction: tell your customers you will never change bank details by email.
«Ten minutes on the phone, on the number you already had: that is what makes the fraud fail.»
Protecting against account takeover, in five layers
The accented layer is detection: the log already exists; what it lacks is a reader.
Identities. Two-step verification asks for a second proof on top of the password; Microsoft wrote in 2019 that it blocks over 99.9 % of account compromise attacks. It is not enough against a relay page, which passes the code along too. Security keys and passkeys, based on the WebAuthn standard, resist better, because the authenticator only offers a credential to the site on which it was registered: on a forged page, it has nothing to give. They go first where an account opens onto money, management, accounts, mail administration; the principle is the one in the guide to sign-in codes and passkeys. Mail. A signed domain with a reject policy; automatic forwarding to outside addresses forbidden by the administrator, which business mail services let you set; a banner that flags external senders.
Detection. Four alerts are enough to start with: sign-in from an unusual country or device, creation of a mailbox rule, creation of a forward, mass sending. They go to a named person who knows what to do; it is part of the routine described in the guide to managed IT. Payments. The six-step procedure, written and signed by management; if the company's bank offers dual signature on transfers, it is turned on. Response. The first-hour sheet, on one page, with the numbers to call, rehearsed once a year.
Closing the door in 2027, in weeks
Four weeks in the example, the first with nothing changed.
- What the manager sees: who has two-step verification, the month's unusual sign-ins, the rules and forwards present in the mailboxes, changes of bank details and who approved them.
- What the company no longer does: change bank details on an email, discover a fraud through the supplier's reminder, believe that a changed password settles everything.
- What is measured: share of accounts with two-step verification, alerts handled and how fast, bank-detail changes checked by phone, the result of the yearly exercise.
- What Symloop does: the stock-take of accounts and mailboxes, two-step verification and keys, mail settings, alerts and their follow-up, the payment procedure written with your accounts team, the exercise, and, on the day it happens, the first hour with you.
Week 1, stock-take: the list of accounts and of those with two-step verification, the rules and forwards of every mailbox, a month of sign-in log, and the look-alike domains of yours already registered by others. It happens that a rule nobody recognises turns up. Week 2, accounts and mail: two-step verification for all, keys for management and accounts, external forwarding forbidden, a reject policy on the domain.
Week 3, payments and alerts: the procedure for changing bank details written and signed, supplier records checked by phone, one by one for the largest, the four alerts set and assigned. Week 4, exercise: a forged message sent to the team, a forged bank-details request sent to accounts, and the first-hour sheet run through once. The exercise is not there to trap anyone: it is there to check that the call back has become a reflex.
- What should we do if the company mailbox has been hacked?
- Seven actions within the hour, in this order: change the password from a clean computer, close every unknown session and device, delete the rules and forwards that were added, turn on two-step verification and check the recovery number, warn the bank if a payment has gone out, warn customers and suppliers by phone, keep the log and the messages. Only then do you look for what was read.
- How do I know whether someone is reading my mailbox?
- From the mail service's sign-in log, which gives dates, places and devices; from the list of rules and forwards on the mailbox; from the list of connected sessions and devices. Messages marked as read that you did not open, or contacts reporting strange messages, are later signs.
- We paid to forged bank details: can the money be recovered?
- Sometimes, if the bank is warned very quickly; the more hours pass, the less likely it is, and this guide can promise nothing. Warn the bank at once, keep the messages with their headers and the log, warn the real supplier, and take the next steps, complaint and recourse, with your legal adviser.
- Is two-step verification enough?
- It stops most attacks that rely on a stolen password; Microsoft speaks of over 99.9 % of account compromise attacks. It does not withstand a forged page that relays the code as well. For management and accounts, add a security key or a passkey, tied to the site, and an alert on unusual sign-ins.
- How do you recognise a forged supplier email?
- Read the full address, not the displayed name: a domain one letter away is the most frequent sign. Be wary of any change of bank details announced by email, of urgency, and of a request not to phone. But no eye is infallible: it is the procedure, the call back on the known number, that protects.
- Should the antivirus not have stopped it?
- No: there is no malicious program on the computer. The person typed their password on a forged page, and the intruder signed in to the mail service from their own device. Two-step verification, sign-in alerts and the payment procedure are what protect.
- Should we change address or domain name after being hacked?
- No. Clean the mailbox, close the sessions, protect the account, and keep the address your contacts know. Changing address would add confusion, and a fraudster could use it to announce your “new address” themselves.
- What should we tell customers and suppliers?
- The truth, by phone first: a mailbox was opened by a third party between such and such dates; accept no change of bank details from us without calling us on the number you already have. What the company owes third parties whose data may have been read is a matter for your legal adviser.
- A message that seems to come from the manager asks for an urgent, confidential transfer: what should we do?
- It is the other form of the same fraud. Do not reply, call the manager on the known number, apply the two-approval rule. An unusual transfer order is never confirmed through the channel that brought it.
- How long does it take to put all this in place?
- Four weeks in the example: one for the stock-take, one for accounts and mail, one for the payment procedure and the alerts, one for the exercise. Two-step verification and the call-back rule can be decided on the first day.