Business Email Compromise
A mailbox is being read and used. Usually about money, always about the audit log.
BEC is the highest-volume real incident most organizations have, and the unified audit log is usually the only place it can be proven. It is also the incident type where the damage is most often financial and ongoing while you investigate, because the attacker is mid-conversation with someone who trusts the mailbox they are writing from.
The first hour
- Establish whether the session is still active, and revoke sessions rather than only resetting the password. A stolen refresh token survives a reset.
- Look for inbox rules first. A rule hiding replies is what lets the fraud continue, and it is usually the earliest attacker action.
- Check for mail forwarding on the mailbox and for org-wide transport rules, which have far greater reach.
- Check whether an OAuth application was granted access. If so, the password was never the point and resetting it changes nothing.
- Identify what was sent from the mailbox. That list decides who else has to be warned, and warning them is often urgent.
Decisions you will be asked to make
What did the attacker actually hold?
| Option | What it costs |
|---|---|
| A password | Reset resolves it, assuming no other foothold. |
| A live token | Reset does nothing. Sessions must be revoked explicitly. |
| An OAuth grant | Neither reset nor session revocation helps. The grant must be revoked and the app removed. |
| Delegated mailbox rights | Survives every credential action. Must be removed as a permission. |
Determine this before acting, because the reflex action resolves exactly one of the four. Evidence of a consent grant with no failed sign-ins is the case people miss most often.
Can you establish what was read?
| Option | What it costs |
|---|---|
| Yes, via MailItemsAccessed | Requires Purview Audit (Premium). Gives a defensible scope. |
| No | Scope must be inferred from access windows, rules and sends, and the report has to say so. |
Check the license before you check the log. Reporting "no evidence messages were read" when the tenant never recorded reads is the single most consequential error in BEC work.
Notify the counterparties now or after scoping?
| Option | What it costs |
|---|---|
| Now | May be premature and alarm people unnecessarily. Stops fraud in progress. |
| After scoping | More accurate. Every hour of delay is another hour the fraud can complete. |
Where money is moving, warn immediately and refine later. Payment fraud completes in hours and is largely irreversible once settled.
Scoping questions
- When was the first successful sign-in that was not the user, and from where?
- Which rules, forwarding addresses or permissions were created, and do any still exist?
- What was sent from the mailbox, and to whom outside the organization?
- Does the tenant record MailItemsAccessed, and for what period?
- Were any other mailboxes accessed with the same session or grant?
Collect, beyond the standard triage set
- Entra sign-in logs for the account, including successful sign-ins with risk detections attached.
- Unified audit log for the mailbox across the full suspected window.
- The inbox rules as they exist now, before removing them.
- Message trace for sent items, and the headers of anything fraudulent.
- The OAuth grants and enterprise applications on the account.
Known mistakes
- Deleting the malicious inbox rule before recording it. The rule is evidence of intent and of what the attacker was hiding.
- Resetting the password and closing the case while an OAuth grant keeps working.
- Scoping to one mailbox when the same grant or token reached several.
- Assuming legacy protocols are disabled. IMAP and POP bypass interactive MFA, and are chosen for exactly that reason.