A phishing message tells an employee to open Microsoft's real device sign-in page.
The domain is right. The password prompt, if there is one, is real. MFA can succeed. And the session on the other end belongs to the attacker.
That's device code phishing in one sentence, and the first half of 2026 turned it from a specialized trick into a packaged workflow.
By September, Microsoft had documented the EvilTokens phishing service (opens in new tab), its affiliate panel, and a coordinated disruption of its underlying infrastructure. The kit's story is a useful field guide to what defenders should watch, even when one service gets knocked offline.
Understand the Handoff
The OAuth device authorization flow has a legitimate job. It lets a device with limited input, such as a conference-room unit, display a short code. The person enters that code in a browser on another device and approves sign-in. The device that started the flow then receives the resulting token.
In a phishing attempt, the attacker starts the flow. The employee receives a pretext and the code, then completes the sign-in on the genuine Microsoft page. The provider authenticates the person; the attacker-controlled client receives the authorized access. MFA did its job of verifying the user. It didn't verify that the user intended to sign in the attacker's device.
The sequence is simple enough to fit on a sticky note:
- Lure: An email, document, chat message, or call claims a file or account action needs attention.
- Code: The attacker-controlled page generates a live code and presents it as the next step.
- Real sign-in: The employee enters the code at Microsoft's authentic device-login page and completes any required MFA.
- Token: The attacker polls for success and receives a valid token for the session they started.
Doppel's device code phishing explainer walks through the protocol and the human decision. Doppelpedia's entry on consent phishing is useful background, too, but the two techniques aren't identical: a classic illicit consent grant asks the user to approve an app's permissions, while device code phishing can hand an attacker tokens through the device authorization flow without a newly registered malicious app.
What Changed in Six Months
Microsoft was tracking Storm-2372's device code campaign (opens in new tab) in February 2025. The newer story is commercialization. EvilTokens appeared in February 2026 and offered campaign components through a phishing-as-a-service panel: lure templates, deployment options, tracking, token handling, and support. Attackers could rent the machinery instead of building every step.
In April, Microsoft described campaigns (opens in new tab) that generated codes dynamically when a victim arrived at a landing page. Older attempts that placed a pre-generated code in an email faced a roughly 15-minute expiry. A fresh code at click time gives the attacker a much wider delivery window. The page can then redirect the victim to the real Microsoft sign-in site.
Microsoft's September 2026 analysis (opens in new tab) put numbers around EvilTokens: 12,000+ inboxes compromised across over 10,000 organizations worldwide. The service charged $1,500 up front and $500 a month for kit and panel access, and it offered dozens of lure themes. Microsoft and partners also disrupted the infrastructure it used to operate. That's an important intervention, but it doesn't retire the underlying device-code flow or the social engineering that abuses it.
If you want the broader economics, Doppel's PhaaS explainer covers why rentable panels keep reappearing, and this Kali365 analysis shows another device-code-flavored kit.
The takeaway: Attackers can swap copy, landing pages, and operators while keeping the same basic handoff.
A kit also shifts the response problem. Blocking one landing page may interrupt one affiliate while other operators keep the panel, templates, and delivery channels. Track related senders, redirect domains, code pages, and token-use patterns together. That gives the SOC a campaign to investigate instead of a pile of unrelated links.
The ‘real URL’ is only the second half of the trip
Some lures include an attacker-controlled page before the Microsoft handoff. Microsoft's April investigation described redirects through compromised domains and familiar cloud-hosted services, plus pages that generated codes on demand. In the September EvilTokens report, the landing page could copy the code to the clipboard and send the employee to microsoft.com/devicelogin.
That means a clean-looking final URL doesn't clear the original message or the page that supplied the code. The employee isn't being asked to decide whether Microsoft owns its own domain. They're being asked to approve a session they didn't initiate. It's a different question, and old "check the URL" training answers the wrong one.
A Field Guide for the SOC
The investigation shouldn't stop at a successful MFA event. Microsoft Entra lets teams filter (opens in new tab) sign-in logs by authentication protocol for device code flow, and Microsoft's threat research describes follow-on activity worth correlating. No single indicator proves compromise; context and sequence matter.
Start with these questions:
- Was the flow expected? Compare the user, client application, resource, time, and purpose with documented device-code use.
- Where did the token go next? Look for unfamiliar IPs, hosting providers, device details, or resource access soon after the code was entered.
- What changed after access? Investigate new inbox rules, Microsoft Graph reconnaissance, mailbox access, device registration, or unusual file activity.
- How did the person get there? Preserve the original email, attachment, URL chain, chat, or call context. The first lure may explain the otherwise legitimate sign-in.
Avoid turning timing into a universal rule. Some operators act quickly; others wait hours before creating a mailbox rule or accessing files. An unusually fast token use can be a clue, not a verdict. The stronger case links an unexpected device-code event to the application, network, and post-authentication behavior around it.
Doppel's GhostCode investigation should remind you that the initial lure can be several steps away from the final authorization. A business contact form, an email thread, and a device-code prompt can all be part of one attack path.
Make the Legitimate Flow Harder to Abuse
For Microsoft Entra environments, the cleanest technical question is whether the organization needs device code flow at all.
Microsoft recommends blocking it wherever possible through Conditional Access (opens in new tab). If a Teams or shared device depends on it, scope exceptions narrowly and test the effect on real workflows. Microsoft's security defaults (opens in new tab) began blocking device code flow for new Entra tenants in July 2026; existing tenants and customized policies still need review.
Then build the human and response layers around the policy:
- Teach the right question: "Did I start sign-in on a device I recognize?" A genuine Microsoft page isn't enough if an unsolicited message supplied the code.
- Rehearse the handoff: Simulate an email-to-code or chat-to-code request and measure whether employees pause, verify, and report it.
- Hunt the follow-on: Correlate device-code sign-ins with unusual token use, mailbox rules, device registration, and reported lures.
- Contain the whole session: If you suspect compromise, revoke refresh tokens and sign-in sessions, inspect active access, remove attacker changes, and follow the identity provider's incident guidance. Microsoft notes that existing access tokens may remain valid for a period after standard revocation.
This is a reminder that authentication and authorization answer different questions. Stronger identity controls, tightly scoped device-code use, and a workforce that knows what a legitimate request looks like all help close the gap.
For incident response, preserve what the employee saw before removing it. The code page, initial message, redirects, app identity, and sign-in log can tell different parts of the same story. If you collect only the final Microsoft URL, the case file may look harmless precisely where the attack succeeded.
Turn Kit Intelligence Into Practice
Doppel's threat graph connects lures, domains, landing pages, and related campaign assets, while its human risk management and simulation turn observed tactics into realistic employee exercises. When the final sign-in page is legitimate, the useful external signal may be the redirector, sender, attachment, or impersonated brand that got the person there.
That makes the training goal more precise. Don't reward someone only for spotting a bad URL. Test whether they recognize an unsolicited device code, verify the request through a known channel, and report the whole sequence. The human decision is the control the kit is trying to borrow.
The Kit Can Change. The Trust Boundary Doesn't
EvilTokens shows how quickly a protocol abuse can become a subscription business, and how much damage can fit inside a real sign-in flow. Its disruption matters. So does the fact that the same legitimate feature remains available where organizations allow it.
Would your employees recognize an attacker-initiated code on a genuine Microsoft page? Request a Doppel demo to see how threat-informed simulations and campaign visibility can help your team test that decision before a live kit does.