Nobody stole a typed password off MadApe’s decoy pages. Recipients still walked Microsoft 365 access tokens, and often refresh tokens, straight to an attacker-controlled device-code client. According to Group-IB, the previously undocumented MadApe kit used business-themed PDF lures and fake SharePoint, OneDrive, Teams, Adobe, and DocuSign pages to deliver an attacker-generated device code, then sent people to Microsoft’s legitimate device login portal so authentication, including MFA, finished on real Microsoft infrastructure. Group-IB published those findings on 29 October 2026 and did not release a count of compromised accounts or treat organization names in the lures as confirmed victims.

How MadApe’s multi-stage lure reached device login

MadApe opened with PDFs dressed as ordinary workplace document traffic. Intermediary pages impersonated familiar file-sharing and e-sign brands so the next click felt like opening a shared file, not approving a foreign OAuth client. Group-IB’s expanded infrastructure set covered 155 root domains, with device-code phishing components on 97 of them, and the kit spoke at least nine languages.

The phishing page showed a device code, copied it when the recipient interacted, reported that interaction to the backend, and handed the person off to Microsoft’s portal. According to Group-IB’s kill-chain write-up, authentication itself took place on legitimate Microsoft infrastructure while the attacker-controlled pages only guided the recipient into completing the device-code flow. No password capture on those decoys appears in the public reporting. The social engineering job was simpler: keep the victim calm long enough to paste the attacker’s code and finish the real sign-in.

Microsoft’s device login page is designed for headless or input-constrained clients. A user reads a code on one screen, opens the device portal on another, pastes the code, and proves they are allowed to sign in. MadApe flipped that convenience. The “device” waiting for the grant was the operator’s client. Everything the victim trusted about the microsoft.com origin stayed true, which is exactly why the trap worked.

Why finishing MFA still issued tokens to attackers

Victims who reached the portal completed authentication, including MFA, on Microsoft’s real pages. Push approvals, app codes, and other transferable second factors still green-lit a session bound to the attacker’s device-code request. The factor proved a person could satisfy the IdP. It did not prove the client holding the pending grant was a company laptop the helpdesk would recognize.

Access tokens followed that approval. Where Microsoft also issued refresh tokens to the attacker-controlled client, the operator kept replayable Microsoft 365 API access after the lure went cold. Group-IB did not publish how many accounts reached that longer-lived stage. The public record also does not name a threat actor group behind MadApe, document ransom demands, or list confirmed victim companies.

Cloud mail, files, and collaboration APIs do not re-challenge the human for every call once a usable token exists. A single successful device-code grant therefore travels farther than a one-off web form theft. Defenders hunting only for harvested passwords on lookalike domains miss the point: the credential ceremony happened on the real IdP, and the wrong party collected the output.

Control in play What MadApe did Why it failed
Password on fake form Not collected Decoys only staged the code handoff
MFA on microsoft.com Victim completed it Approval bound to attacker client
Device-code grant Attacker generated the code User pasted it into real portal
Refresh token (when issued) Stayed with attacker client Replay needed no second lure

OTP seeds, SMS messages, and approve/deny prompts remain portable secrets or portable decisions. Anyone coaching a paste-and-approve flow can ride them when the IdP still offers a device-code path that mints tokens for a client the user never enrolled as a trusted workplace device.

What Group-IB could not confirm from the kit alone

Group-IB treats names inside lure content as intended targets, not proven compromises. Confirmed account counts, record volumes, and financial loss figures are absent from the disclosure. Login mechanics beyond device-code approval on the real Microsoft portal are also unpublished. Those gaps matter for board metrics. They do not soften the mechanism: a multi-stage kit turned everyday document bait into a clean token handoff without ever hosting a convincing password field.

Similar device-code abuse has shown up in other Microsoft 365 social-engineering waves because the flow is official, mobile-friendly, and easy to narrate over chat or email. MadApe’s contribution was packaging. PDF themes, brand impersonation, multilingual copy, and dozens of domains lowered the friction between “open this file” and “paste this code.”

Stopping the next wave is not a matter of training people to distrust microsoft.com. The portal is genuine. The failure sits earlier, in identity policy that still allows a remote stranger’s pending grant to receive a full user authentication. Device-bound, phishing-proof authentication binds the factor to the device and the origin, so it cannot refuse a genuine Microsoft page while the tenant still allows the grant. The control that removes the handoff is closing device-code grants at the corporate IdP.

The defensive side of this case is covered in the related article on mfa2point0.com: how blocking the device-code grant stops MadApe-style Microsoft 365 token theft.

FAQ

How did the MadApe kit get Microsoft 365 access without stealing passwords?

MadApe got Microsoft 365 access by steering recipients to Microsoft’s legitimate device login portal with an attacker-generated device code. According to Group-IB, victims pasted that code and finished authentication, including MFA, on real Microsoft infrastructure, after which access tokens and, where issued, refresh tokens went to the attacker’s client instead of a corporate device.

Did MFA protect accounts targeted by MadApe device-code phishing?

MFA did not protect accounts in the MadApe device-code flow when victims completed the real Microsoft prompts for the attacker’s pending grant. The second factor confirmed the person could authenticate. It did not stop Microsoft from issuing API tokens to the operator’s device-code client after that approval.

What infrastructure did Group-IB tie to MadApe?

Group-IB tied MadApe to an expanded set of 155 root domains, with device-code phishing components on 97 of them, and reported support for at least nine languages. The firm described PDF lures plus intermediary pages impersonating SharePoint, OneDrive, Teams, Adobe, DocuSign, and similar brands leading into the Microsoft device portal.

Were specific companies confirmed as MadApe victims?

Specific companies were not confirmed as MadApe victims in Group-IB’s public findings. Organization names appearing in lures were treated as intended targets, and no compromised-account count, ransom figure, or named operator group was published with the 29 October 2026 disclosure.

Why do refresh tokens matter after a MadApe-style lure?

Refresh tokens matter after a MadApe-style lure because, when Microsoft issues them to the attacker-controlled device-code client, the operator can keep calling Microsoft 365 APIs without sending another PDF. Access tokens already unlock cloud workloads for their lifetime; refresh material extends that window until revocation or expiry catches up.