In the first months of 2026 two stories about Microsoft 365 phishing ran side by side. On 4 March, Microsoft and Europol disrupted Tycoon 2FA, the phishing-as-a-service kit that let thousands of criminals get past multi-factor authentication. A few weeks earlier, a new service called EvilTokens had started selling a different trick: instead of stealing a password and an MFA code, it convinced people to type a short code into a genuine Microsoft page and hand over a working session. Looking back from November, the lesson of both is the same, and it is one that most companies have not yet acted on: "we have MFA" is no longer the end of the conversation. This article explains both techniques in plain terms and lists the settings and habits that actually stop them. For the broader picture, start from our guide to phishing protection.
Key Takeaways
- Both techniques steal the session, not the password. Adversary-in-the-middle (AiTM) kits relay your login in real time; device code phishing gets you to approve the attacker's sign-in yourself.
- Codes and push notifications do not stop either. Phishing-resistant MFA (FIDO2 keys, passkeys, Windows Hello for Business) defeats AiTM, but not device code phishing, because the victim signs in on the real Microsoft domain.
- The decisive control for device code phishing is configuration: block the device code flow with Conditional Access, keeping a short, documented list of exceptions.
- Takedowns buy days, not safety. CrowdStrike saw Tycoon 2FA traffic drop to 25% of its earlier level on 4 and 5 March 2026, then return to previous levels.
- Users need one new rule: a code you did not ask for, on a device in front of you, is never yours to enter.
What happened in spring 2026
On 4 March 2026, under a court order from the U.S. District Court for the Southern District of New York, Microsoft and Europol, with industry partners, seized 330 domains that ran Tycoon 2FA's control panels and fake login pages, as reported by Decipher. The figures released with the action show its scale: tens of millions of phishing emails a month reaching more than 500,000 organisations, about 62% of the phishing attempts Microsoft blocked by mid-2025, and around 96,000 victims since 2023, more than 55,000 of them Microsoft customers. The kit targeted Microsoft 365, Outlook and Gmail accounts and captured credentials and session tokens in real time, which is what let its users bypass MFA (Security Affairs).
The disruption did not last. In a post dated 20 March, CrowdStrike reported that daily volumes fell to 25% of their earlier level on 4 and 5 March and then returned to pre-disruption levels, with tactics unchanged.
In parallel, a research note published by the Cloud Security Alliance on 25 March documented EvilTokens, a phishing-as-a-service platform launched on Telegram on 16 February 2026. Its infrastructure was first seen on 19 February, and by mid-March it had reached more than 340 Microsoft 365 organisations in the United States, Canada, Australia, New Zealand and Germany, in sectors from construction and manufacturing to healthcare, legal services and local government. EvilTokens did not proxy logins. It used device codes.
Two techniques, one goal: your session
Adversary-in-the-middle (AiTM)
The victim receives a link, often behind a CAPTCHA or a QR code, and lands on a page that looks exactly like the Microsoft or Google sign-in. That page is a reverse proxy: it forwards everything to the real service and sends the real answers back. The user types the password, receives the push or the code, approves it, and is logged in. So is the attacker, who has kept the session cookie that the real service issued. From then on, the attacker does not need the password or the second factor.
Device code phishing
The device code flow is a legitimate part of OAuth 2.0 (RFC 8628), built for devices without a keyboard, such as smart TVs, meeting-room systems and command-line tools. The device shows a short code; you open a sign-in page on another device, type the code, and the TV is logged in. Attackers turn this around, as the CSA note describes:
- The attacker requests a code from Microsoft's device authorisation endpoint. The code typically expires in about fifteen minutes, so the lure is sent live.
- The victim is asked, by email, chat or phone, to enter the code at
microsoft.com/devicelogin, presented as a one-time code to open a document, hear a voicemail or sign a contract. - The victim completes the normal MFA challenge on genuine Microsoft infrastructure.
- The attacker's software, which has been polling Microsoft in the background, receives an access token and a refresh token.
Refresh tokens can last weeks or months and survive a password reset; revoking them takes an explicit administrative action. The CSA note lists what followed in observed cases: reading mail, OneDrive, contacts and calendars through Microsoft Graph, and registering attacker-controlled devices in Entra ID.
| AiTM phishing | Device code phishing | |
|---|---|---|
| Where the victim types | A fake page that relays to the real one | The real Microsoft page |
| What the attacker obtains | Session cookie (and the password) | Access and refresh tokens |
| Stopped by codes or push MFA? | No | No |
| Stopped by FIDO2 or passkeys? | Yes, the key refuses the wrong domain | No, the domain is genuine |
| Main configuration defence | Phishing-resistant MFA, compliant devices | Block the device code flow |
Why "we have MFA" is not the answer
An SMS code, an authenticator code or a push notification proves that the right person is present. It does not prove where that person is typing. That is why AiTM works against all of them. Phishing-resistant methods, such as FIDO2 security keys, passkeys and Windows Hello for Business, bind the sign-in to the real domain, so a proxy on a look-alike domain gets nothing usable.
Device code phishing is different. The victim is on the real domain and passes the real MFA, so the origin check succeeds. The CSA note is explicit that FIDO2 does not prevent it. What stops it is removing the flow for the people who do not need it.
None of this makes MFA less necessary. The 2026 Verizon DBIR finds credential abuse somewhere in the attack path of 39% of breaches. MFA closes the cheap attacks; the techniques above are what is left, and they are now sold as a service.
Configuration defences in Microsoft 365
These settings live in the Microsoft Entra admin center. Conditional Access needs Entra ID P1 licensing, which is included in Microsoft 365 Business Premium and in the E3 and E5 plans.
- Block the device code flow. Microsoft's own guidance is to "get as close as possible to a unilateral block". Create a Conditional Access policy for all users and all resources, set the Authentication flows condition to Device code flow and grant Block access, as described in Microsoft Learn. Run it in report-only mode first, exclude your break-glass accounts, and allow the flow only for documented cases such as meeting-room devices or legacy tools that cannot be updated.
- Consider blocking authentication transfer (moving a sign-in from a PC to a phone) with the same condition, if you do not use it.
- Move administrators to phishing-resistant MFA first, then extend it to finance, HR and executives, the roles that receive the most targeted lures.
- Require a compliant or joined device for sensitive applications. A token stolen to an attacker's laptop is much less useful when access requires a managed device.
- Block legacy authentication, which ignores MFA entirely.
- Require admin consent for applications. Users should not be able to grant mailbox access to unverified third-party apps.
- Disable automatic external forwarding and review inbox rules, the first thing attackers create once inside.
On Google Workspace, the equivalent priorities are security keys or passkeys for administrators and control over which third-party apps can access account data.
What users must learn
Most awareness training says "check the address bar". Against device code phishing that advice fails, because the address bar shows microsoft.com. Training needs one new, simple rule:
A legitimate device code starts from a device in front of you, such as a TV, a meeting-room screen or a tool you just launched. A code that arrives by email, Teams, WhatsApp or phone is never yours to enter.
Useful points to add to the next session:
- The pretexts seen in 2026 were ordinary: construction bid requests, DocuSign notifications, voicemail alerts, Microsoft Forms prompts and shared-file reminders.
- Lures increasingly arrive outside email. The DBIR 2026 reports that in phishing simulations, the median click rate through mobile channels such as SMS and voice is 40% higher than through email.
- "Approve this sign-in" or "enter this code" while on the phone with "IT support" is the moment to hang up and call back on a known number.
- Reporting fast matters more than never clicking. A report in the first minutes lets IT revoke the session before mail is read.
Detection signals for IT and the SOC
- Device code sign-ins. In Entra ID sign-in logs, look for events whose authentication protocol is
deviceCode. If your organisation does not use the flow, any such event is a strong indicator. - Unusual origin. Sign-ins from hosting providers, Tor exit nodes or commercial VPNs shortly after a user reported a strange code request.
- Fast data access. Bulk reads of mail or OneDrive through Microsoft Graph within minutes of the sign-in.
- New devices. A device registered in Entra ID soon after a device code sign-in.
- Mailbox changes. New forwarding or deletion rules, new delegations, new OAuth consents.
If you confirm a compromise, revoke the user's sessions and refresh tokens, reset credentials, and review devices registered, rules created and consents granted after the suspicious sign-in. The CSA note suggests reviewing at least 90 days of sign-in logs to find earlier cases.
Where Orizon fits
Orizon Aware sends each role the phishing it is likely to receive, teaches what was missed right after a click and measures who reports, who clicks and who improves. Adding the device code rule above to your training content is the cheapest control on this list.
RECON Cloud connects to Microsoft 365 and Google Workspace with one admin consent and read-only Graph access, and checks MFA and 2-Step Verification for admins, conditional access, legacy authentication, guest and consent settings, risky or unverified OAuth apps, and external mail forwarding and delegation. Treat the device code policy as a separate line on your checklist and confirm it in the Entra admin center. For detection, Oversight, Orizon's managed SOC on the CYREBRO platform, collects identity and email sources, so a Microsoft 365 login with stolen credentials or a new forwarding rule is read together with everything else.
Checklist
- Audit device code usage in the sign-in logs of the last 90 days.
- Create the block policy in report-only mode, review the impact, then enforce it.
- Document every exception, with an owner and a review date.
- Enforce phishing-resistant MFA for administrators.
- Block legacy authentication and require admin consent for apps.
- Disable automatic external forwarding.
- Add the "unrequested code" rule to awareness training and to your phishing simulations.
- Write a short playbook: revoke sessions, check devices, rules and consents.
If you want to see how your Microsoft 365 or Google Workspace tenant is configured today, and whether your people would recognise a code lure, start with RECON Cloud for the settings and Orizon Aware for the people. Neither replaces the Conditional Access policy above: that one is yours to switch on.