← All articles

Microsoft 365 Phishing Abuses Device Code Sign-ins

Microsoft 365 Phishing Abuses Device Code Sign-ins

Microsoft 365 phishing campaigns are abusing legitimate device-code authentication, a sign-in method that lets someone authorize a device by entering a short code, to compromise US corporate accounts. ANY.RUN documented the activity in sandbox investigations, which run suspicious files or websites in an isolated environment so researchers can observe their behaviour safely. The campaigns mix trusted online services, real authentication pages and familiar business documents, increasing the risks of account takeover, data theft, financial loss and operational disruption.

SkyW4lker, an active phishing kit, uses Microsoft's legitimate device-code process rather than directly stealing a password or bypassing multifactor authentication (MFA), the extra identity check commonly performed through an app, code or security device. ANY.RUN identified 618 related sandbox analyses between June and October 21, 2026, and said 46% came from the US. Phishing emails presented as contracts, invoices or business proposals send victims through CAPTCHA checks, which are tests intended to distinguish people from automated software, and trusted platforms. Victims then reach the legitimate Microsoft authentication page, where entering the attacker's device code can authorize access to their account.

During the SkyW4lker investigation, ANY.RUN used its Related Threat Context capability to connect one suspicious address with earlier sandbox sessions and broader threat intelligence. Researchers moved from a SkyW4lker session into the company's TI Lookup service and examined IP address 198.23.223.185. That work exposed related URLs, malicious infrastructure and technical connections to another phishing kit called Gabagool. The linked activity involved technology, government, education, consulting and healthcare organisations, with submissions originating in the US, Latin America and Europe. Security operations centre (SOC) teams, which monitor and respond to digital threats, can add the associated URLs and indicators of compromise (IOCs), observable signs of malicious activity, to detection and threat-hunting systems.

A second group of campaigns exploits business interest in artificial intelligence. ANY.RUN linked 1,258 sandbox analyses to fake applications and event invitations impersonating ChatGPT, Claude and other popular services. The US ranked second worldwide for submissions overall and first in two major campaign groups. Fraudulent downloads delivered malware, while deceptive event invitations attempted to capture access to Microsoft 365 and Google accounts. In one case, a genuine Claude installer appeared to run normally while concealed processes copied browser session cookies, data that can preserve a signed-in session. Those cookies could potentially let an attacker access business accounts without causing another MFA prompt.

GeorgeGinx is an active Phishing-as-a-Service (PhaaS) operation, meaning criminals provide ready-made phishing infrastructure for others to use. It targets US government, banking, healthcare and IT organisations. The operation places redirects behind legitimate Adobe Shared Links and sends victims to Microsoft device-code phishing pages, potentially authorizing corporate-account access without disclosing a password. ANY.RUN examined HTML Document Object Model (DOM) changes, the browser's internal representation of a web page, along with HTTP requests, the messages exchanged between a browser and web servers. This revealed concealed destinations behind the Adobe links and an administration panel on an attack server. The finding underlines why a familiar document-sharing domain alone does not establish that a link is safe.

WKLP is a Python-based information stealer and remote access trojan (RAT), malware that lets an attacker control an infected computer from elsewhere. It has been observed targeting government organisations in the US and France. WKLP steals credentials, browser information and sensitive files while maintaining access to compromised systems. It executes malicious code through PowerShell and a legitimate Python runtime, adds Microsoft Defender exclusions, terminates security processes and creates firewall rules disguised to avoid attention. ANY.RUN detected malicious Python activity presented as MoUsoCoreWorker.exe. These changes weaken endpoint protection and can support extended access and data theft.

Mir0Auth is another phishing kit targeting Microsoft 365 users, with significant activity directed at US organisations. Its emails carry malicious PDF attachments that redirect recipients to fake DocuSign pages. The campaign abuses legitimate Device Code authentication to obtain access tokens, digital values that prove a user is already authorized, without first requiring a stolen password. ANY.RUN reported finding obfuscated scripts, meaning code deliberately made difficult to read, and hidden authentication requests used to collect Microsoft 365 tokens. A stolen token could potentially expose sensitive business messages and data.

Defensive measures differ slightly by campaign. Organisations can restrict unnecessary device-code authentication through Microsoft Entra Conditional Access, monitor suspicious device-code activity and revoke compromised Microsoft 365 sessions. Administrators should approve trusted AI applications, prevent unauthorized software installation, verify unexpected event invitations and inspect redirects rather than trusting a sharing service's domain. For WKLP, the recommended controls include Microsoft Defender Tamper Protection and monitoring for unauthorized Defender exclusions, unusual PowerShell use, terminated security tools and unexpected firewall changes. ANY.RUN also recommends keeping detection systems supplied with recently observed malicious URLs and domains.

ANY.RUN says its tools can make threat triage 94% faster, reduce mean time to respond (MTTR) by 21 minutes per incident and cut escalations from Tier 1 to Tier 2 analysts by 30%. It also says 30% of malicious IOCs obtained from customer sandbox investigations are unique, allowing organisations to add infrastructure that other intelligence sources may not contain to their existing security systems. These are company-reported performance claims. For organisations seeking help with the wider security and infrastructure issues raised by these attacks, AEU-I provides security-first IT, infrastructure and consulting.

Also available in: DeutschItalianoShqip