MGB
MGB // Cyber Blog
← All incident files

Microsoft: a forgotten test account

A password spray against a legacy, non-production test account gave a nation-state actor a foothold in Microsoft's own corporate email. It took more than six weeks to detect, and what the attacker read in the mailboxes led to more.

The short version
  • Entry point: a password spray compromised a legacy non-production test tenant account, which gave the actor a foothold.
  • Reach: the account's permissions let the actor read a very small percentage of Microsoft corporate mailboxes, including senior leadership and staff in cybersecurity and legal. The actor exfiltrated some emails and attachments.
  • Detection: Microsoft detected the attack on 12 January 2024, more than six weeks after it began in late November 2023.
  • Second order: in March 2024 Microsoft said the actor used information from the stolen email to try to reach source code repositories and internal systems, and that some customer secrets had been shared with Microsoft by email.
Late Nov 2023Attack begins
12 Jan 2024Detected
10xPeak spray increase, Feb 2024 vs Jan
0Evidence of customer-facing systems compromised, as of Mar 2024

Timeline

Went wrong Worked Context

Late Nov 2023

The password spray succeeds. The actor compromises a legacy non-production test tenant account and gains a foothold. Microsoft gives no exact date.

Following weeks

The actor reads corporate email. Using the account's permissions, they access a very small percentage of Microsoft corporate mailboxes, including senior leadership and staff in cybersecurity, legal and other functions. They exfiltrate some emails and attachments, initially looking for information about the actor itself.

12 Jan 2024

Microsoft detects the attack. It activates its response process to investigate, disrupt, mitigate and deny further access.

19 Jan 2024

Public disclosure. Microsoft says it has no evidence, to date, that the actor accessed customer environments, production systems, source code or AI systems.

Feb 2024

Password spraying surges. Microsoft sees spray volumes rise as much as tenfold compared with the already high January volume.

8 Mar 2024

The attacker uses what it stole. Microsoft says the actor used information from the exfiltrated email to gain, or try to gain, access to some source code repositories and internal systems. Some secrets shared between customers and Microsoft over email were in the stolen mail, and Microsoft starts contacting affected customers. It still finds no evidence that Microsoft-hosted customer-facing systems were compromised.

What went wrong

  • A forgotten account was a way in. A legacy non-production test account could be password-sprayed, and it carried enough permission to read corporate mailboxes.
  • Dwell time. From late November to 12 January is more than six weeks of access before detection.
  • Email became an intelligence source. The attacker read mail to find out what Microsoft knew about it, then used what it found to go further.
  • Secrets in email. Credentials and other secrets shared by email with customers were in the stolen mail. That turned a Microsoft breach into a potential problem for those customers.

What worked

  • Detection and response. Microsoft detected the activity on 12 January and activated its response process straight away.
  • Scope held, on the evidence given. As of March 2024 Microsoft reported no evidence that Microsoft-hosted customer-facing systems had been compromised.
  • Public updates. Microsoft published on 19 January and again on 8 March, and said it would apply current security standards to legacy systems and internal business processes even where that disrupts existing processes.

What would have helped

This section is MGB's own analysis, not part of the source material.

  • Inventory and retire test and legacy accounts. Every account and test tenant should have an owner and an expiry. If nobody owns it, disable it.
  • Hold every account to the same baseline. Conditional Access, multifactor authentication and blocking of legacy authentication should cover test, service and break-glass accounts too, with a documented exception route rather than a quiet gap.
  • Watch for password spray. Alert on many failed sign-ins across many accounts from the same source, and on successful sign-ins to accounts that rarely sign in, using Entra sign-in logs and Identity Protection.
  • Do not email secrets. Use a secure sharing route, and rotate any credential that has ever been sent by email.
  • Treat a supplier's breach as a trigger. If a vendor says its mail was accessed, rotate anything you ever sent them, without waiting to be told it was in scope.
Not in the sources Microsoft's two posts do not say whether multifactor authentication was enabled on the compromised account, so this file does not either.