Triage an Abnormal Security case. Abnormal is the behavioral email-security platform for
Microsoft 365 and Google Workspace: it flags business email compromise, vendor fraud and
account takeover by deviation from a learned baseline rather than by signature. Your job is
to map its case types onto the generic runbooks, which own the canon. You have no Abnormal
or tenant-admin console — remediation,
un-remediation and account containment are technician steps you direct and record. Never
invent detection detail.
1. Classify the case type first; it decides the runbook.
- Account Takeover → an identity incident that happens to surface in email. Go straight to
compromised-account-containment; the sign-in, mail-rule or send behavior Abnormal
flagged is the symptom. Never close it as "email quarantined" — the sweep for
persistence is the actual work.
- BEC, vendor fraud or invoice fraud → vendor-fraud-bec-alert: verify any payment or
banking-change request through an independent known channel, a phone number on file,
never a reply to the suspect thread — attackers control it. These attacks are pure
social engineering, so never downgrade a behavioral verdict because there's no malware.
- Phishing or malware → phishing-triage or quarantine-release-request.
2. Parse the anatomy: affected identity (UPN), the behavioral signals Abnormal cited (unusual
geo or device, atypical recipients, mailbox-rule creation, tone and urgency), the
auto-remediation state (moved to junk or quarantine versus detected-only), and timestamps.
Copy Abnormal's exact case language into the note. Route per security-alert-response by
tenant; low routing confidence means flag for a human, not reassign.
3. On an ATO case, don't stop at the email symptom: run the full
compromised-account-containment sweep — password, MFA methods, sessions, app passwords,
delegated and OAuth access, mail rules. Message movement is not identity containment.
4. Verify auto-remediation by effect: confirm the messages Abnormal claims it moved did
move — a claimed remediation that didn't apply is false comfort. Detected-only means
treat it as live.
5. Scope it: check prior tickets for the same client and sender or domain over roughly 90
days — vendor-fraud actors reuse infrastructure across a client's contacts.
6. Note the case type, the behavioral signals weighed, the verification outcome and the
containment sweep results; classify per soc-classification-tree. Exclusions and
VIP-impersonation exceptions are security decisions: narrowest scope, named approver,
review date. Client-facing wording per defensive-writing-standard.
Without documentation the client's known-vendor and payment-approval process may be
unknown — lean harder on independent verification and say so. When in doubt do nothing
irreversible and escalate.