Skip to content
high severityMicrosoft Sentinel

Detect MFA fatigue attacks in SigninLogs (KQL)

MFA fatigue (also called push bombing or MFA request generation) is when an attacker who already has a valid password triggers repeated MFA prompts, hoping the user eventually approves one. This SigninLogs query counts sign-ins where MFA was required but not satisfied — result codes 50074 and 500121 — and flags accounts hit with several prompts in a short window.

Pair it with a follow-up check for a later successful sign-in from the same IP to catch the moment the user gives in and approves.

The query

SigninLogs
| where TimeGenerated > ago(1h)
| where ResultType in ("50074", "500121")   // MFA required / MFA denied or not completed
| summarize
    MfaPrompts = count(),
    FirstPrompt = min(TimeGenerated),
    LastPrompt = max(TimeGenerated),
    Apps = make_set(AppDisplayName, 20)
    by UserPrincipalName, IPAddress, bin(TimeGenerated, 30m)
| where MfaPrompts >= 5
| sort by MfaPrompts desc

Line-by-line explanation

  • | where ResultType in ("50074", "500121")

    50074 means strong authentication was required but not provided; 500121 means the MFA request failed, was denied or timed out. Together they capture prompts the user did not approve.

  • by UserPrincipalName, IPAddress, bin(TimeGenerated, 30m)

    Groups per account, source IP and half-hour so a burst of prompts to one user collapses into a single row.

  • | where MfaPrompts >= 5

    Flags accounts that received five or more unsatisfied prompts in the window — well above a user mistyping once or twice.

Required tables and columns

  • ResultTypeResult code; 50074 and 500121 indicate MFA was required but not completed.
  • UserPrincipalNameThe account being prompted; the subject of a fatigue attack.
  • IPAddressSource IP of the sign-in attempts generating the prompts.
  • AppDisplayNameApplication that triggered the MFA challenge.

MITRE ATT&CK mapping

Tuning and false positives

  • Add a follow-on query for ResultType 0 from the same UserPrincipalName and IPAddress just after the burst to find prompts the user eventually approved.
  • Raise the threshold for help-desk or shared accounts that legitimately re-authenticate often.

Limitations

  • An attacker who paces prompts slowly stays under the per-window count; widen the bin to catch low-and-slow campaigns.
  • Number-matching and push-limit protections change which codes appear; verify the codes your tenant emits before relying on exact values.

References

Last reviewed 2026-11-18.