Why a callback policy fails inside a real invoice thread
A callback policy fails when a bank change arrives as a reply in a real vendor thread. Here is why staff skip the call, and how to make the callback automatic.

In our reading, a callback policy gets skipped inside a real invoice thread because the thread already looks like proof. The request comes from the vendor's real mailbox, passes every authentication check, and sits under weeks of genuine messages, so the call starts to feel like doubting a colleague rather than checking a stranger.
The fix is a callback that fires on the content of the request: any change to bank details or payment instructions gets a logged call, and the payment system refuses the change until that call is recorded.
Why a real invoice thread makes the callback policy feel optional
A hijacked vendor mailbox sends the bank change from inside a real conversation. The Irish National Cyber Security Centre names this case "Supplier fraud: A vendor's email is compromised and used to send fake invoices or change payment details"1.
The attacker does not need to invent the context. The NHS Counter Fraud Authority says fraudsters "monitor, hack, or spoof (copy) email accounts to impersonate a known and trusted entity" and "gain their trust by using genuine information about the employee or supplier"2. MITRE ATT&CK notes that a compromised email account "may engender a level of trust in a potential victim"3. For how attackers build the lure in the first place, see AI spearphishing and thread hijacking.
What the thread proves, and what it does not
A hijacked reply carries five genuine signals, and each proves less than it appears to.
| Signal in the thread | What it proves | What it does not prove |
|---|---|---|
| Sent from the vendor's real address | The message came from that mailbox | That the vendor's staff wrote it |
| SPF, DKIM and DMARC pass | The vendor's domain sent it | That the account was not taken over |
| Weeks of earlier messages | The relationship is real | That the newest message is |
| Correct invoice numbers and names | The sender can read the thread | That the sender is the account manager |
| No first contact safety tip | The sender has written to you before | That this request is normal |
The last row comes from how the email tool works. Microsoft's first contact safety tip appears "The first time they get a message from a sender" or when "They don't often get messages from the sender"4. A vendor who mails your finance team every month meets neither condition, so the tip would not be expected to appear on the hijacked reply. This is our reading of how the setting is described, not a Microsoft statement.
Where the callback policy breaks in practice
The callback policy breaks at three points: the trigger, the number and the approver.
The trigger is a person's judgment. The Irish NCSC lists the red flags staff are trained to spot: "urgency, secrecy, external domains, unusual tone"1. A hijacked reply in a known thread can avoid all four, so nothing prompts the call.
The number and the approver are the other two. A number taken from the email reaches the attacker, and an approver who also read the email is not a second check; see "How to make the callback policy part of the payment system", steps 2 and 3.
The FBI's 2025 Internet Crime Report lists business email compromise losses of $3,046,598,558, second only to investment fraud5. The Association for Financial Professionals found that "About three in four organizations (74%) were affected by business email compromise (BEC) in 2025"6.
How to make the callback policy part of the payment system
Three changes to the payment process and two staff steps make the call happen without anyone deciding to make it.
- CHANGE. In your accounts payable procedure for payment changes (owned by the finance controller), rewrite the callback trigger so it reads on the content: every request to change bank details or payment instructions gets a callback, whatever thread it arrives in and whoever sent it. A trigger left to the reader's judgment lets a real-looking thread skip the call. The Irish NCSC's wording is "Use a strict call-back verification process for any change in bank details"1.
- CHANGE. In your ERP or accounts payable system, open the approval workflow for vendor master changes (the menu name differs by product) and require a second approver for any change to a vendor's bank account, with fields for who made the callback, to which number and when. The Irish NCSC asks for dual approval of "financial transactions over a set threshold"1. Applying a second approver to every bank-detail change, whatever the amount, is our recommendation, because that one change decides where every later payment goes. The system then holds the payment even when the person reading the thread is convinced.
- CHANGE. In the same vendor record, put the callback phone number under the same approval as the bank details, so it cannot be edited from the change request. Staff call only that number, as it stood before the request (where the system keeps a change history, it shows the earlier value). The NHS Counter Fraud Authority says "the supplier's contact details should be taken from existing records held by the NHS organisation and not from information supplied in the change request"2. The number then stays out of reach of an attacker who controls the thread.
- STAFF. Brief accounts payable in onboarding and at each month-end close that a reply in a real vendor thread is the expected shape of this fraud, the case NCSC Ireland calls "Supplier fraud"1, so it gets the call like any other change. Use the call steps in how to check a payment change request as the script.
- STAFF. Have the finance controller or internal audit send a planned bank-change request inside an existing thread each quarter, and count from the callback fields in the vendor change log ("How to make the callback policy part of the payment system", step 2) how many requests reach approval without a logged call. The NHS Counter Fraud Authority suggests "assessing a finance team's response to a fictitious supplier request (e.g. phone, email, video call) to make a bank account change"2. A test that uses a fresh, suspicious email only measures the easy case.
What to ask on the callback
The callback only works if it asks something the thread cannot answer. An attacker inside the vendor's mailbox can read every message in it. This is our reasoning from how thread hijacking works, not a rule from a standard.
Use this as the script for the call in "How to make the callback policy part of the payment system", step 1. Ask the contact on the known number to confirm that they sent the change, and to state the new account details from their own side before you read anything out. Then ask about something that was never in email between you, such as a detail agreed on an earlier call. If the person cannot answer, or offers to call back from a different number, hold the change.
Avoid the questions the thread answers: the invoice number, the amount due, the name of the account manager.
Why email banners will not flag this reply
Email warnings are built around the sender, and the sender here is real. Microsoft's external sender tag, turned on with Set-ExternalInOutlook -Enabled $true7, marks every message from outside your organisation, except senders on its allow list. Every message in a vendor thread carries the same tag, so it does not separate the fraudulent reply from the genuine ones.
When a mailbox on your own side is taken over, the attacker can also set a rule that copies payment threads to an outside address; how to find malicious forwarding rules covers that check.
Catch thread-hijacked payment changes with AegisAI
"AegisAI checks each payment request against what the sender has sent before", and "Its verdict rests on what this sender has sent before and on the request itself"8. For a bank change in a known thread, that sender history is the deciding signal.
"Every verdict arrives with the evidence, so an analyst can confirm it in seconds"8. AegisAI does not make the call or approve the change; those stay in your payment process. Keep the callback policy and the second approver whatever email security you run. To see a payment request checked against its sender's history, book a demo.
Frequently asked questions
Why do staff skip the callback policy when the email is in a real thread?
In our reading, because the thread shows the signs they were trained to trust: a known sender, passing authentication and real history.
Does DMARC stop a bank change request from a hijacked vendor mailbox?
No. The vendor's own domain sends the message, so SPF, DKIM and DMARC pass. DMARC checks which domain sent the message, not whether the account owner wrote it.
Which phone number should the callback use?
The number in your vendor record as it stood before the request, as in "How to make the callback policy part of the payment system", step 3.
Is a phone call enough on its own?
No. Add the second approver from "How to make the callback policy part of the payment system", step 2.
What should we ask on the callback?
Ask the contact to state the new account details from their side, and ask about something that was never in email.
Sources
Each number in the text opens the source with the same number.
- 1National Cyber Security Centre (Ireland)Protecting Against Business Email Compromise and Payment Fraud, 25 July 2025
- 2NHS Counter Fraud AuthorityPayment diversion fraud: guidance for prevention and detection, July 2026
- 3MITRE ATT&CKCompromise Accounts: Email Accounts, Sub-technique T1586.002
- 4Microsoft LearnAnti-phishing policies in Microsoft 365
- 5FBI Internet Crime Complaint Center2025 IC3 Annual Report
- 6Association for Financial ProfessionalsOver 75% of US Firms Experienced Payments Fraud in 2025, 14 April 2026
- 7Microsoft LearnSet-ExternalInOutlook (ExchangePowerShell)
- 8AegisAICheck a payment change request email: call-back rules
.png)