AegisAI Perspectives · 116
Email security that catches BEC with no link or attachment reads the sender relationship and the request, plus account behavior, even when DMARC passes.
See it on your own Google Workspace or Microsoft 365 tenant. No MX change.
Or get the Q3 2026 State of the AI Attack report (free, instant download)
AegisAI agents check every payment request against the sender’s history17. The email security that catches BEC with no link or attachment reads the sender relationship and the request, then the payment wording and the account's behavior, since a lookalike domain can pass email authentication5. BEC losses reported to the FBI reached $3,046,598,558 in 20251. Score every tool you consider on those four points.
Put these three outcomes at the top of your buying sheet.
They have nothing to score, because the message is plain text that asks for a payment or a reply9. Microsoft describes spoofed messages that trick a recipient into replying with sensitive content as business email compromise. In Microsoft 365, a composite authentication failure does not by itself block a message; the service weighs it with the message's overall suspicious nature8.
The FBI's 2025 report also counts more than $30 million in business losses to BEC scams involving AI3. The comparison of AI models and rules for payloadless BEC covers which method reads which signal, so this page stays on what to ask when you buy. Drop any tool from the shortlist that judges these messages only by their links and files, or by their authentication result.
Short on time? Watch the explainer, or book a demo.
A tool has to know whether this sender and this recipient have written to each other before, and about what6. Microsoft 365 shows a first contact safety tip the first time a recipient gets a message from a sender, or when they rarely get mail from that sender. Its mailbox intelligence flags a sender who impersonates an address in the user's personal sender map11.
Mailbox intelligence protection does not work if the sender and recipient have previously communicated by email7. Ask every vendor where its sender history comes from, and whether it still applies when the sender is a known contact.
A tool has to read what the message asks the reader to do: pay, change bank details, reply with data or keep it quiet. Microsoft lists large language model content analysis as one of the detection technologies on its email entity page10. Gmail's spoofing settings check whether a sender's name matches a name in your Google Workspace directory12.
The FBI advises verifying any change in account information through a secondary channel or two-factor authentication4. Ask each vendor to show a verdict on a plain text bank change request, with the reason it gives.
A tool has to notice when a genuine mailbox starts asking for things its owner never asks for. A hijacked vendor mailbox signs as itself, so SPF, DKIM and DMARC pass18. Gmail's unauthenticated email setting requires a message to be authenticated by SPF or DKIM13, which a hijacked mailbox already meets.
Ask each vendor what its tool does when a trusted contact's real address sends a first ever request to change payment details.
Run the same eight steps on every tool, on your own tenant, and record the result of each.
List the built-in settings you already run: the first contact safety tip, mailbox intelligence, plus Gmail's employee name and unauthenticated email settings.
Connect each tool in a mode that leaves mail routing as it is, and note whether it asks for an MX change.
Send a finance user a plain text request to change a supplier's bank details from a test domain you control that has valid SPF and DKIM records.
From a test mailbox that has written to that user before, send a request that moves a payment to a new account. This tests the gap where mailbox intelligence stops.
Send a plain text request for a reply with staff details. It carries no link and no file.
Record for each message the verdict, the reason shown, the minutes to verdict and the action taken: warning, spam, quarantine or removal14.
Count real supplier messages the tool flagged in the same week, from its own log.
Check that your callback rule ran for every bank change, whatever the tool decided.
Put these eight steps in the trial plan before the first vendor call.
Score the built-in controls and any added tool on the rows below, so the gap an added tool has to fill is visible.
| Criterion | Built-in Microsoft 365 or Google Workspace controls | AegisAI |
|---|---|---|
| Sender relationship | First contact safety tip and a personal sender map per user | Every payment request checked against the sender’s history |
| Known contact, new request | Mailbox intelligence does not work after prior contact | Reads the sender’s history and the change being asked for19 |
| Request wording | Large language model content analysis is a listed detection technology | Reads where the reply actually goes |
| Authentication passes | A lookalike domain with valid records passes | Stops the BEC that authentication lets through |
| Action after a verdict | Gmail warns, moves to spam or quarantines | Removed from every mailbox that received it20 |
| Bank change check | Your team confirms it by a secondary channel | No edge. The callback on a number on file stays with your team |
| Mail routing | No change | Monitoring mode on live mail, no MX change21 |
The bank change row gives no tool the edge, because the FBI's secondary channel check stays a human step. Score each row yes or no for every tool, and use the demo request page to have a plain text bank change scored on your own mail.
AI agents check every payment request against the sender’s history, and stop the business email compromise that authentication lets through. AegisAI reads the thread the way an analyst would: the sender’s history, the change being asked for, and where the reply actually goes. The median verdict takes 1.9 seconds, then the email is removed from every mailbox that received it. You can run it in monitoring mode on live mail with no MX change, then turn on removal.
Illustrative example, figures are fictional. Northfield Labs (fictional) has 180 staff on Google Workspace and one IT lead. On a Monday at 10:12, the accounts payable clerk gets a plain text email from the real mailbox of a supplier the company has paid for 3 years. It asks to move this month's 2 invoices, worth $48,000, to a new bank account before Thursday. The message passes SPF and DKIM, so the unauthenticated email setting does nothing. Tool A flags it at 10:13 and names the new bank account as the reason. Tool B lets it through, because the sender is a known contact. The clerk calls the supplier on the number on file, as the callback rule requires, and the supplier confirms it sent nothing.
Run this case past each vendor and write down the minute its tool would have raised a finding.
Track three numbers from the first day of the trial.
Review all three on day 30 and keep a tool only if the first two fell and the third reached every change.
One that reads the sender relationship and the request together with the payment wording and the account's behavior. Authentication alone is not enough, because a lookalike domain with valid DNS records passes it. Test it on a plain text bank change request before you buy.
No. A lookalike domain with valid DNS records passes authentication, and a hijacked vendor mailbox signs as itself, so SPF, DKIM and DMARC pass. Keep DMARC, and add a tool that reads the request.
They catch part of it. Microsoft 365 shows a first contact safety tip and checks a personal sender map, and Gmail checks sender names against your directory. Mailbox intelligence does not work once sender and recipient have written before.
The FBI's IC3 recorded BEC losses of $3,046,598,558 in 2025 and $2,770,151,146 in 2024. In 2025 businesses also reported more than $30 million in losses to BEC scams involving AI.
Verify it through a secondary channel or two-factor authentication before anything changes, as the FBI advises. Call the requester on a number already on file, never one in the email.
