The "Booking Reference" on a fake meeting scheduler was a live Microsoft device code. Every page the victim checked was either convincing or genuinely Microsoft's, so there was no bad URL to spot.
Key Takeaways
Most phishing advice comes down to one habit: check the address bar before you sign in. Device code phishing is built to pass that test. In a case Huntress published on November 12, an employee at a construction company did exactly what the training says to do and ended up on a real Microsoft page, on a real Microsoft domain, entering a code that handed an attacker a working session to their account. Nothing about the final step was fake. The deception lived entirely in what the victim believed the code was for.
The attack began in mid-October with a spear-phishing email from a typosquatted domain, registered eight days earlier, impersonating a senior vice president at a different construction company. The pitch was an "Introductory call" about a long-term supply partnership, the kind of message a sales or procurement employee would want to answer. The link was wrapped in Microsoft Safe Links and pointed at a Google short link, and Huntress found the chain behind it ran three redirects server-side at the same instant, which let it slip past URL reputation checks.
The landing page looked like a Microsoft Teams calendar setup, complete with the impersonated executive's headshot and name. The target chose a meeting length, date and time and pressed Continue. Behind the scenes, the page's backend started a genuine Microsoft device code sign-in and received a user code in return. The "Confirm Your Booking" screen then displayed that code as a "Booking Reference," with a session countdown to add urgency. Clicking Copy or Confirm Meeting quietly placed the code on the clipboard and opened Microsoft's real device sign-in page in a popup. The victim pasted the code, approved the request, and the popup closed on a "Verification complete" message before the browser moved on to an ordinary Calendly booking page. From the user's side, they had just scheduled a call.
The kit was built to recover from the obvious failure points. If Microsoft responded with error AADSTS50076 or interaction_required, typically a sign that MFA was needed, the page showed a "verify your account again" message and quietly started a fresh device code session. A separate "WinHello" stage displayed a fake Microsoft prompt asking for additional verification on the device. The kit itself was stored as AES-256-GCM ciphertext and decrypted in the browser, so email scanners and crawlers that do not run JavaScript saw only an encrypted blob.
Device code flow exists so that a smart TV, printer or command-line tool without a keyboard can borrow a browser on another device to sign in. The device shows a code, a person types it at Microsoft's verification page, and the device receives tokens. The flaw in that design is that Microsoft cannot know whose device is waiting for the code. When an attacker starts the request, the victim's approval gives the attacker's machine the session, and because the victim completes whatever sign-in steps are required on the genuine page, the attacker never needs a password and, in many configurations, never faces an MFA challenge of their own.
In this case, Huntress saw the attacker's follow-on sign-in to the Microsoft Authentication Broker, logged as Cmsi:Cmsi, routed through a residential proxy in New Orleans so it would resemble consumer traffic. Its analysis notes the account had no MFA enrolled and was reached from a non-compliant, unmanaged device, and that the session was replayed from attacker infrastructure with the ASN and region changing inside a single session ID. Huntress contained the intrusion before it escalated, but the same access in a mailbox with no one watching would have been an ideal starting point for invoice fraud.
That is not hypothetical. BleepingComputer reported in September that Microsoft disrupted EvilTokens, a device code phishing service run by a group it tracks as Storm-2992, which compromised more than 12,000 inboxes across over 10,000 organizations. Phishing data recovered by SpyCloud showed 8,708 compromised accounts across 6,585 corporate email domains in 79 countries. The service sold for $500 a month or $1,500 outright with 44 customizable kits, and after a takeover it used Microsoft Graph to map relationships and search mailboxes for wire transfer details and pending invoices before generating tailored business email compromise messages. The report stresses the operation was not a full takedown, and affiliates have already built clones such as APToken.
The trend predates both cases. Proofpoint, as reported by ITPro last December, called the shift toward abusing trusted authentication flows "a major evolution in phishing," tracking a high-volume criminal actor, TA2723, and a Russia-linked cluster it calls UNK_AcademicFlare, alongside red team tools such as SquarePhish that make the technique easy to reproduce.
The most direct fix is one most tenants have never configured. Microsoft's Conditional Access documentation recommends organizations "get as close as possible to a unilateral block on device code flow," allowing it only in well-documented, secured cases such as legacy tooling that cannot be updated. The policy is a few clicks: target all users and all resources, set the Authentication Flows condition to device code flow, grant Block access, and run it in report-only mode first to see what legitimately depends on it. Emergency access accounts should be excluded and that exclusion list audited regularly.
Detection matters for whatever remains allowed. Huntress points to Cmsi:Cmsi entries as evidence that a device code sign-in succeeded, to programmatic user agents such as python-requests and node that often appear only in non-interactive sign-in logs, and to ASN or region changes within one session ID. After a confirmed compromise, Huntress advises checking for rogue MFA methods, unrecognized registered devices, unauthorized app consents, suspicious mailbox rules and suspicious outbound mail, the footholds an attacker plants so that access outlasts the first session.
The construction company's employee checked the domain and found it was Microsoft's, because it was. That is the lesson of this campaign: when the final page is genuine, the defense has to sit in the tenant's policy rather than in the user's judgment, and the policy Microsoft recommends is already a toggle away for most organizations.
Guide
Like the reverse-proxy kits this guide dissects, the Teams booking lure let the victim finish a genuine Microsoft sign-in while the attacker walked away with the session. Its audit of which MFA factors can still be intercepted is the right first step after this case.
Download
Whitepaper
Huntress contained the device code takeover at this construction company before it escalated. This paper explains how round-the-clock monitoring of Microsoft 365 gives firms without their own SOC a chance to stop account takeover before it becomes invoice fraud.
Download
Guide
A spear-phishing email from a lookalike domain started this attack, and checking the final URL could not stop it. This guide's layered model, pairing email controls and awareness with detection that catches a successful phish before it spreads, fits a lure built to pass the address-bar test.
Download
One opened message file leaks an NTLMv2 hash, and a queue sorted by severity reaches the fix last.
The attacker signed in as the backup app itself, so user MFA and Conditional Access never applied.
Organizations blind to stolen session cookies had a 50% identity event rate, against 37% for those that monitor them.