The employee finished MFA on Microsoft’s genuine sign-in page. Tokens still landed with the attacker, not with the person holding the phone. According to QMasters DFIR, device-code phishing persuaded the user to type a short code into that real Microsoft page, which approved a login for an attacker-controlled session without stealing the password, and the victim’s own MFA did not stop the grant. Roughly 1,600 messages and shared files sat in the exposure window across 19 days, then supplier impersonation and forged-invoice fraud followed from the same mailbox.

How a short code on Microsoft’s real page opened the account

Device-code OAuth is built for keyboards that cannot comfortably finish a full interactive login. The attacker starts the flow, receives a user code, and coaches the target to enter that code at Microsoft’s legitimate sign-in experience. The user sees familiar branding, completes whatever MFA prompt their tenant requires, and believes they have authenticated themselves. The authorization server issues tokens to the party that started the device-code session. That party is the attacker.

QMasters reports this path as the most likely initial access for the case. The original lure that first put the code in front of the employee was not recovered, and Entra sign-in history retention on the licensing in use was only seven days, so the first login itself was not still visible when investigators arrived. A later attacker lure recovered from the mailbox matched the device-code pattern when sandboxed. The mechanics still show up clearly in what followed the first successful grant.

Within minutes of first token issuance, an unrecognized Windows device was registered to the account. Investigators later counted 963 device-code flow logins from that device against the single compromised identity. MFA completion on the genuine page had already done its job for the wrong principal. Password theft was unnecessary. Cookie scraping from malware was unnecessary. The grant ceremony itself handed the session over.

Whatever prompt the employee satisfied authorized the attacker’s device-code session on Microsoft’s real origin. That gap does not change the outcome. Whatever prompt the employee satisfied, it authorized the attacker’s device-code session on Microsoft’s real origin.

Control in play What happened Why it failed the defender
Employee MFA prompt Completed on real Microsoft page Approved attacker-started device-code grant
Password secrecy Password never needed from victim OAuth device flow issued tokens anyway
New Windows device Registered minutes after first token Became a foothold separate from the password

What nineteen days of that session actually bought

Across a 19-day Microsoft 365 compromise window, the attacker used the issued session to reach about 1,600 messages and shared files. Finance threads, password notes, and OTP messages were among the material opened. QMasters notes that bulk file download versus preview could not be fully settled from available telemetry, and no download was recorded in the attacker session they could attribute. Access alone was enough to run the business email compromise playbook.

On day 9 the attacker opened a real supplier invoice thread and saved documents to OneDrive. A lookalike domain was registered. A draft reply was deleted. In the days after, supplier impersonation with lookalike domains and a forged invoice pushed payment fraud. Discovery did not come from an internal alert on device-code volume or a strange registered device. It came externally, on the morning of day 20, when the supplier forwarded suspicious correspondence.

After first password reset and token revocation, the attacker session continued for roughly an hour or more through the still-valid access token and the registered-device foothold. QMasters puts the operational lesson in plain language: a reset is not an eviction. Revocation ends refresh tokens, but an access token already issued can keep working for up to about an hour, and a registered device is a separate foothold that has to be disabled or deleted on its own. During that residual window the attacker read internal warnings, launched outbound phishing toward well over a hundred addresses (more than two hundred sharing invitations and phishing messages in the reporting), and created inbox rules meant to hide the trail.

A second revocation, account block, device deletion, and removal of an unverified publisher app finally cut the path. Residual rules and sharing links stayed active briefly. The unverified app had been added to the organization the day after initial access; who approved it is unknown, and the source does not tie that approval to the attacker.

Password notes sitting in the compromised mailbox named shared-mailbox, ERP, and social credentials spanning related companies. Those notes most likely exposed a Google Workspace shared-mailbox password with super-admin rights at the related firm. That password remained unchanged for up to about 18 days. QMasters found no evidence it was used, and none of 224 attacker-tied addresses appeared in the related company’s Google logs. Offline secrets in mail are not an interactive login problem, but they turned one M365 session into cross-cloud credential exposure on paper.

Available CrowdStrike data showed no endpoint malware indicators for the window investigators could see. Full disk collection was not available; the employee PC was reset before imaging, and full activity depth was limited. Fourteen Microsoft 365 sign-in devices without matching CrowdStrike computer names went unchecked in the data on hand. The case that is documented is still a cloud-identity and token story, not a confirmed workstation implant story.

Why “MFA worked” still left finance mail open

Defenders often treat a completed MFA prompt as proof the right human won the login. Device-code phishing breaks that assumption without spoofing the IdP page. The user is on the real Microsoft origin. The second factor fires. The authorization still binds to the attacker’s waiting client because that is how the device-code grant is designed to finish when a human pastes the code.

That is why password resets and refresh-token revocation felt incomplete here. The first failure was the remote grant that should never have issued an unbound attacker session after a coached code entry. Everything after that, the registered Windows device, the lingering access token, the invoice thread, the lookalike domain, the outbound phishing after discovery, rode tokens and footholds that already existed. Cleaning those up is eviction work. It is slower, noisier, and easy to under-scope when someone assumes “we reset the password” equals “they are out.”

High device-code volume from a device the user does not recognize, a new registered device minutes after first token, finance mail plus password notes, and supplier lookalike domains are the practical signals. The pattern is not. High device-code volume from a device the user does not recognize, a new registered device minutes after first token, finance mail plus password notes in the same mailbox, and supplier threads that suddenly sprout lookalike domains are the practical signals. After the incident, QMasters recommended detection for device-code sign-ins and for newly registered devices.

A single Microsoft 365 mailbox still concentrated invoice truth, OTP mail, and password notes that pointed at other firms. One coached code was enough to sit in that concentration for 19 days.

The defensive side of this case is covered in the related article on mfa2point0.com: how to prevent Microsoft 365 device-code phishing after MFA.

FAQ

How did the attackers get into the Microsoft 365 account?

The attackers most likely got in through device-code phishing. According to QMasters DFIR, the employee entered a short code on Microsoft’s genuine sign-in page and completed MFA, which issued tokens to an attacker-controlled session without the password being stolen. An unrecognized Windows device was registered to the account within minutes of the first token.

Did MFA fail in the QMasters device-code BEC case?

MFA prompted and the employee completed it on the real Microsoft page, yet that completion still authorized the attacker’s device-code grant. The QMasters write-up states the victim’s own MFA did not stop token issuance to the attacker. The failure was the remote device-code approval path, not a missing prompt.

Why did a password reset not kick the attacker out immediately?

A password reset and refresh-token revocation in this QMasters IR case left a still-valid access token and a registered device usable for roughly an hour or more. QMasters documents that revocation ends refresh tokens while an already-issued access token can keep working for up to about an hour, and that a registered device must be disabled or deleted on its own. During that window the attacker read internal warnings and sent outbound phishing.

What did the attacker access during the 19-day window?

QMasters DFIR reports about 1,600 messages and shared files in the exposure window, including finance material, password notes, and OTP messages. The attacker used a real supplier invoice thread, lookalike domains, and a forged invoice for payment fraud, then phished contacts from the mailbox after discovery. Password notes also most likely exposed a related company’s Google Workspace shared-mailbox super-admin password for up to about 18 days, with no evidence of use.

Was endpoint malware required for this BEC?

Available CrowdStrike telemetry in the QMasters case did not show endpoint malware indicators for the window investigators could review. The documented path centers on device-code token issuance, a registered Windows device, and continued cloud session use, not a confirmed workstation implant. Full disk imaging was unavailable because the employee PC was reset before collection.