This sheet is past its review date. Deadlines or advice may have changed since 5 Nov 2026.
Incident response: what do you do in the first 24 hours?
This month, confirm that Microsoft 365 auditing is on, keep sign-in logs for at least a year and agree who leads, who decides and whom you call. When something looks wrong, isolate accounts and devices without wiping them, export the logs in the first hours and, if the GDPR requires it, notify within 72 hours1.
If you only do five things
Check that the audit log is on and reaches back far enough.
Run
Get-AdminAuditLogConfigin Exchange Online PowerShell: auditing isn't on by default for Business Basic, Standard or Premium2. Audit (Standard) keeps records for 180 days3, but Entra ID keeps sign-in logs for only 7 days on Free and 30 days on P1 or P2, so route them to an Azure storage account4 or your SIEM.Line up outside help before you need it.
Sign an incident response retainer, or find out which firm your cyber insurer sends and what it needs from you first. The NCSC advises UK organizations of any size to use an assured provider, and notes that insurance may include one5; in Germany, the BSI lists qualified APT response firms6.
Decide now who may pull the plug.
Name who can disconnect or shut down systems, call in outside help and speak publicly, with deputies for nights and holidays. NIST lists that authority in the incident response policy9, and the NCSC warns that critical response roles should never rely on one person7.
Rehearse a stolen mailbox once a year.
Business email compromise complaints to the FBI reported $3.05 billion in losses in 202510. Set up and rehearse procedures to revoke sessions, isolate devices, reset credentials and check related accounts11. The NCSC's free Exercise in a Box includes phishing scenarios12.
The first hours
Name a lead and a note-taker.
Record what was found and when, and each decision with who made it and why, in one time zone13. A paper logbook is fine9.
Can you trust email and Teams?
If an admin account or a mailbox may be compromised, assume the attacker can read them and move to your other channel. Communicating over the network under investigation is a common mistake14.
Isolate first, and don't wipe.
Disconnecting a system preserves evidence that shutting it down can lose, though shutting down may halt active processes13. Defender for Endpoint, including Plan 1, isolates a device but keeps it connected to Defender, and lifts isolation after seven days15. Stop backup jobs and protect the backups from further changes16.
Export the logs now.
Search the audit logs from just before the first suspicious activity, without filtering by activity at first17. On Entra ID Free, sign-ins older than 7 days are already gone4. Keep the original phishing email with its headers8.
Call your responders, insurer and lawyer before big changes.
Let the investigation run before you restore systems: the NCSC warns that rushing recovery can significantly increase risk13.
One account or a wider intrusion?
Lock a single compromised account at once (see below). If the attacker has admin rights, plan one coordinated cleanup: partial remediation often tips the attacker off8, and they likely hold other credentials14.
Which clocks have started?
The GDPR's 72 hours start once you are reasonably certain that personal data was compromised, and you can notify in phases1. For NIS2, DORA and the CRA, see the EU security laws sheet. In the UK, a report to the NCSC doesn't meet your legal duties13. To report the crime in Germany, contact your state police's cybercrime contact point (ZAC)18.
If a Microsoft 365 account is compromised
- Disable the account and revoke sessions. In the Entra admin center, clear Account enabled and select Revoke sessions, or run
Revoke-MgUserSignInSession. Apps that aren't CAE-capable keep access until their access token expires, by default after 1 hour19. - Reset the password through another channel. Don't email it to the user. Reset synced accounts twice in Active Directory, and replace app passwords, which a reset doesn't revoke17.
- Remove what the attacker added. Check MFA methods and devices, admin roles and consented apps17. Disable a malicious app rather than delete it, or another user's consent can bring it back20.
- Check forwarding and inbox rules. Run
Get-InboxRule -Mailbox <user> -IncludeHiddenand check the mailbox's forwarding addresses. Rules that forward mail or move it to Notes, Junk Email or RSS Subscriptions are typical signs17. - Find out what was read. MailItemsAccessed is logged by default for E3 and E5 users. If the attacker synced mail to their own client, Microsoft says to assume the entire mailbox is compromised21.
- Warn the people it wrote to. Check Sent Items and message trace for changed bank details and phone the recipients. If money left, ask your bank at once to recall it; the FBI says time is of the essence10.
What changed in 2026
- Voice phishing went mainstream in Teams. Microsoft saw confirmed malicious voice phishing over external Teams calls rise 502% in a year; once a victim grants remote access, attackers finish reconnaissance in two minutes or less11. Mandiant ranks voice phishing second among initial vectors, at 11%22. If someone gave a caller remote access, treat it as an incident and isolate the device.
- Hand-offs take seconds. Mandiant measured a median of 22 seconds in 2025 between initial access and hand-off to a second group, down from more than 8 hours in 2022, and advises treating routine malware alerts as high priority22.
- More intrusions are found in-house, but they last longer. Organizations found 52% of Mandiant's 2025 intrusions themselves, up from 43%, while median dwell time rose to 14 days from 11. Mandiant calls standard 90-day log retention insufficient22.
- The NCSC published a first-hours checklist. Its July 2026 guidance on disruptive attacks covers incident command, outside responders, the disconnect-or-shut-down decision and a central record13. Check your plan against it.
What you can skip
- Rewriting your plan to NIST's new model. SP 800-61 Rev. 3 maps incident response to the CSF 2.0 functions, but NIST says to use whichever life cycle model suits you9. Names, numbers and authority matter more.
- Buying tools during the incident. Microsoft sees tools bought under pressure go unused and advises deferring any you can't deploy during the investigation8.
- Imaging every machine. Analyze the systems the attacker used or changed8. NIST notes that most malware incidents won't be prosecuted, so formal chain of custody isn't always used9.
- Naming the attacker. Attribution rarely changes the first day's actions. Spend the time on which accounts, devices and data are involved.
Mistakes that cost the most
- Wiping before collecting evidence. It loses memory and other host artifacts, and the attacker may notice and change tactics14. You may then be unable to show what was taken.
- Tipping off the attacker. Don't upload samples to online scanners, which attackers watch8, or ping and browse their servers14.
- Fixing only the symptom. Blocking an IP address without learning what the malware is and how it got in lets the attacker keep access14.
- Resetting every password at once. Reset the accounts you know are compromised first, then others in stages8.
- Waiting for the forensic report to notify. Start the short initial investigation at once; detail can follow in a phased notification1.
What this looks like at your size
Small team, no security staff
- Put your IT provider's incident duties and what it may do on your behalf in the contract9, with a response time.
- Hang up the BSI's IT emergency card with your IT contact's number: staff report at once and act only on instruction23.
- Note your insurer's incident number and keep the policy details offline5.
Growing, with IT or security staff
Enterprise or regulated
- Set up incident command, such as Gold, Silver and Bronze13, with delegated authority out of hours7.
- With E5 or an add-on4, Audit (Premium) keeps Entra, Exchange, OneDrive and SharePoint records for 1 year by default3.
- Give each reporting clock an owner and draft the notices in advance, using the EU security laws sheet.