Skip to content
Unlisted Report logoUnlisted ReportSubscribe
News

Microsoft Takes Down EvilTokens AI Phishing Service

The platform used an AI chatbot to plan fraud from 12,000+ hacked inboxes. Microsoft seized 50 sites and UK police arrested two men.

Microsoft and partners have disrupted EvilTokens, a phishing service that used AI to help criminals pick victims and plan fraud, Microsoft says.

What EvilTokens did

  • Sold on Telegram for a $1,500 joining fee and $500 a month, The Record reports.
  • Stole login tokens for Microsoft 365, bypassing multi-factor authentication.
  • Its AI chatbot read victims' inboxes, spotted trusted relationships and payment approvals, and drafted messages impersonating trusted contacts.

Since launching in February 2026, it was linked to more than 12,000 compromised inboxes at over 10,000 organizations. About 1,000 criminals used it.

How the phishing technique worked

EvilTokens was built around "device-code phishing," a technique that abuses the OAuth 2.0 device authorization flow — a legitimate sign-in method originally designed for devices without easy keyboard input, such as smart TVs or printers. Attackers trick victims into entering a legitimate-looking device code on Microsoft's real sign-in page, which then hands the attacker an authenticated session token for the victim's account without the attacker ever needing the victim's password. Because this method generates a real, valid session using Microsoft's own infrastructure, it can bypass multi-factor authentication entirely, since the MFA challenge is effectively satisfied by the victim during the legitimate-looking sign-in step. EvilTokens was first documented as this type of phishing-as-a-service kit by security firm Huntress in March 2026 (The Hacker News).

Microsoft tracks the threat actors behind the platform's development and support under the designation Storm-2992 (The Hacker News).

Scale and reach

Microsoft said it observed the highest concentrations of victim activity in the United States, Canada, the United Kingdom, Australia, India and France, with affected organizations ranging across wholesale distribution, construction, financial services, real estate, higher education and healthcare (Microsoft On the Issues). Microsoft VP of security research Tanmay Ganacharya told The Register that, at the height of the service's use, the company was tracking "10 to 15 distinct campaigns launching every 24 hours" (The Register).

The takedown

  • Under a court order from the U.S. District Court for the Eastern District of Virginia, Microsoft seized 50 websites and disabled more than 150 related domains (The Hacker News).
  • The legal action was brought together with Health-ISAC, a health-sector cybersecurity nonprofit, alongside support from Cloudflare, Coinbase, OpenAI, Railway, SpyCloud, The Shadowserver Foundation and TRM Labs.
  • Cloudflare removed the attackers' infrastructure from its network.
  • London's Metropolitan Police arrested two men, aged 32 and 38, on September 11 or 18, 2026 (reports on the exact date vary slightly), on suspicion of fraud and money laundering offences connected to the alleged operation of EvilTokens. Officers executed warrants at two locations. Both men have since been released on bail while the investigation continues (The Record).

Microsoft Digital Crimes Unit's associate general counsel, Steven Masada, said: "While EvilTokens helped cybercriminals access email accounts, at the center of the service was an AI-style chatbot that could analyze a victim's inbox and help criminals identify trusted relationships, payment authorizations, and sensitive responsibilities, as well as other circumstances where fraud was most likely to succeed… The platform could even recommend fraud strategies, including drafting messages that impersonated trusted contacts to help criminals trick victims into taking action" (The Hacker News).

A Microsoft spokesperson told Recorded Future News: "The two individuals were alleged operators of EvilTokens. We believe others may have supported the service in various capacities" (The Record).

Why it matters

AI is no longer just writing better phishing emails. Here it decided who to target and how to steal the most money — a sign of where business email fraud is heading. As TechRadar put it, EvilTokens operated "like a startup" complete with subscriptions, dashboards and AI-driven targeting, illustrating how professionalized and commercialized cybercrime-as-a-service platforms have become (TechRadar).

Even with the takedown, experts caution that dismantling the infrastructure and arresting alleged operators does not guarantee the underlying technique disappears — other criminal groups could adopt similar AI-assisted device-code phishing methods, especially since the core OAuth abuse technique is not new, only its AI-driven targeting layer.

How to protect your organization

  • Use phishing-resistant sign-in such as passkeys or security keys.
  • Require a phone call to a known number before changing payment details.
  • Review mailbox rules and sign-in alerts for unusual activity.
  • Restrict or closely monitor use of the OAuth device authorization ("device code") flow within your organization, since this is the specific mechanism EvilTokens abused to bypass MFA.
  • Train staff to be suspicious of unsolicited prompts to enter a "device code" on a Microsoft sign-in page, especially ones arriving via email, chat, or text rather than something they initiated themselves.

What readers should do

  1. Be wary of device-code sign-in prompts you did not initiate. If you receive a code and are asked to enter it on a Microsoft sign-in page, and you did not just try to sign in to a new device, treat it as a phishing attempt.
  2. Check your account's active sessions and sign-in activity in Microsoft 365 or your organization's identity portal if you suspect you may have fallen for this type of scam, and revoke any sessions you do not recognize.
  3. Report suspicious emails that reference your real business relationships (vendors, payment approvals, colleagues) to your IT or security team immediately — these personalized details are the hallmark of this kind of AI-assisted fraud.
  4. Verify payment changes by phone, using a number you already have on file, not one provided in the suspicious message itself, before acting on any request to change bank details.
  5. If you believe you were a victim, Microsoft and partner organizations said they were in the process of notifying affected account holders; watch for official notifications and contact your IT department or bank if you suspect fraudulent activity.

Device-code phishing did not start with EvilTokens

EvilTokens industrialized a technique that state-sponsored hackers had already been using for roughly two years. Microsoft first publicly documented device-code phishing in February 2025, attributing a campaign running since August 2024 to a cluster it calls Storm-2372, which it assessed with moderate confidence aligned with Russian interests (Microsoft Security Blog). Microsoft has since updated that assessment, concluding that Storm-2372 is an initial-access sub-cluster of Midnight Blizzard, the same Russian state-linked group separately reported to have used Anthropic's Claude to rebuild its malware (Microsoft Security Blog).

Around the same time, security firm Volexity documented multiple Russian threat actors impersonating officials from the US State Department, Ukraine's Ministry of Defence and the European Parliament, luring targets into fake Microsoft Teams meetings or chatroom invitations that ultimately required victims to complete a device-code sign-in step (Volexity). Because the technique abuses a legitimate, standards-based authentication flow — defined in Internet standard RFC 8628 — rather than a flaw in Microsoft's code, it does not trigger the same red flags as a fake login page, making it unusually hard for users to spot (Cloud Security Alliance research note).

By March 2026, researchers at the Cloud Security Alliance reported that OAuth device-code phishing — by then already commoditized by EvilTokens as a subscription product — had hit more than 340 Microsoft 365 organizations across five countries within weeks of the platform's February 2026 launch (Cloud Security Alliance).

Why MFA does not stop this attack

The Cloud Security Alliance's research note makes a point worth repeating clearly: "Multifactor authentication provides no protection against this attack class: the victim completes the MFA challenge themselves on behalf of the attacker, and the resulting refresh tokens persist even after a password reset, complicating remediation" (Cloud Security Alliance). In other words, standard advice to "just enable MFA" does not fully neutralize this specific threat — organizations need to go further by restricting or disabling the device-code authentication flow itself through Conditional Access policies where it is not genuinely needed, a mitigation also recommended by cybersecurity firm eSentire (eSentire).

From espionage tool to commercial product

The jump from a nation-state tradecraft technique to a $1,500-joining-fee criminal subscription service illustrates a recurring pattern in cybercrime: sophisticated methods pioneered by well-resourced state actors tend to filter down into the wider criminal ecosystem once they prove effective, usually packaged into easier-to-use tools sold on underground marketplaces or, as in this case, Telegram. That trajectory — from a handful of Russian state hackers in 2024 to over 12,000 compromised inboxes sold as a service by 2026 — is part of why Microsoft and its partners treated dismantling EvilTokens' infrastructure as a priority even though the underlying OAuth flow abuse is not itself new.

By · Published · Updated · 10 min read

Sources & Verification

Additional reporting

Read next