The Takedown
On September 22, 2026, Microsoft's Digital Crimes Unit (DCU) announced it had dismantled EvilTokens, an AI-powered phishing-as-a-service platform that had compromised more than 12,000 email inboxes across over 10,000 organizations worldwide since launching in February 2026 (Microsoft, Sept. 22, 2026). Acting on a temporary restraining order from the U.S. District Court for the Eastern District of Virginia in Microsoft Corp. and Health-ISAC, Inc. v. Felix Utomi, Waidi Segun Adams, and Does 1–5 (Case No. 1:26-cv-3047, filed Sept. 22, 2026), Microsoft seized 50 websites and disabled more than 150 additional domains tied to the operation's infrastructure (Microsoft Notice of Pleadings). Ten days earlier, on September 11, London's Metropolitan Police cybercrime team arrested two men, aged 32 and 38, on suspicion of fraud and money laundering; both were released on bail pending further investigation (The Record).
This was the DCU's 40th court-authorized disruption in nearly two decades of using civil litigation against cybercrime networks — and, by Microsoft's own account, its first against an end-to-end, AI-enabled criminal service (The Record).
How EvilTokens Worked
EvilTokens sold access to a dashboard and AI chatbot for $1,500 up front plus $500 a month, marketed through Telegram channels to buyers who wanted to break into corporate inboxes and run business email compromise scams. The technique exploited Microsoft's own device-code login flow: victims who clicked a phishing link were shown a short-lived authentication code and coaxed into entering it on the genuine Microsoft sign-in page, handing over live account access without ever surrendering a password (The Record). Once inside, the AI chatbot did the reconnaissance that used to take a human days — summarizing inboxes, mapping organizational hierarchies, flagging trusted relationships, and surfacing live conversations about pending payments — compressing that work into hours (Microsoft blog). Wholesale distribution, construction, financial services, real estate, higher education, and healthcare organizations were hit hardest, concentrated in the US, Canada, UK, Australia, India, and France (Microsoft blog).
Private Litigation as De Facto Cyber Enforcement
What's notable here isn't just the AI angle — it's the mechanism. No federal cybercrime statute specifically targets phishing-as-a-service platforms as such; Microsoft instead brought its own civil suit alleging unauthorized computer intrusion, unlawful deception, and trademark violations, then used the resulting court order to compel domain registries to transfer or disable infrastructure. Cloudflare, Coinbase, OpenAI, Railway, SpyCloud, the Shadowserver Foundation, and TRM Labs all assisted (Microsoft blog). Congress has not passed comprehensive legislation governing AI-enabled fraud; the FTC's enforcement authority under Section 5 reaches deceptive practices but not infrastructure seizure. In that vacuum, Microsoft's EDVA playbook — refined since its earliest botnet suits in 2010 — has become the fastest available tool, and it is company lawyers, not regulators, wielding it.
The Case for Caution
The strongest objection to this model isn't hypothetical. In 2014, Microsoft used an ex parte court order to seize 23 domains from No-IP.com in a botnet takedown — and in doing so knocked out DNS service for roughly 5 million legitimate subdomains belonging to innocent No-IP customers, to disrupt an estimated 18,000 malicious nodes. The Electronic Frontier Foundation called the action roughly 99.6% overbroad, comparing it to "seizing a whole shopping mall to stop a single mafia-run storefront" inside it. Microsoft reversed course within two days and later settled, acknowledging No-IP had not knowingly supported the malware (EFF, 2014). EFF's core complaint was procedural: No-IP, an innocent third-party intermediary, was given no advance notice before losing control of its business — Microsoft's stated fear that notice would tip off criminals didn't apply to the platform being seized from.
That history matters because ex parte relief is precisely what happened again this month: Microsoft and Health-ISAC obtained their TRO before EvilTokens' operators or any hosting intermediaries had a chance to respond, with defendants given 21 days to answer only after the seizure was already executed (Microsoft Notice of Pleadings). Legal scholars Asaf Lubin and Ido Marinotti have separately argued this pattern amounts to unaccountable "privateering" — private companies exercising quasi-law-enforcement power through courts with limited capacity to test the technical claims underlying a seizure request. Microsoft's own counsel have rebutted this directly, arguing cybercriminals are simply "strategic competitors" using illegal means, no different in kind from trademark or trade-secret theft, and that federal judges routinely handle internet-infrastructure disputes competently (Lawfare). Both sides have a point: civil litigation is fast and available today, while a purpose-built statutory framework for AI-enabled fraud does not exist and would take years to legislate.
A Proportionate Path Forward
The EvilTokens case is a genuinely good outcome — a criminal service that compromised 12,000 inboxes in seven months is gone, and two operators face UK prosecution. Congress shouldn't respond by restricting private civil enforcement; it remains the only mechanism moving at the speed cybercrime demands, and there is no serious floor of legislative or regulatory capacity ready to replace it. But the No-IP precedent shows the gap worth closing isn't authority — it's process. A modest fix: require plaintiffs seeking ex parte infrastructure-seizure orders to identify and notify any innocent third-party intermediary (a registrar, a hosting provider, a legitimate SaaS platform whose login flow was abused, as Microsoft's own was here) within 48 hours of seizure, with an expedited hearing available on request. That preserves the surprise needed against the actual wrongdoer while giving bystanders the due process EFF rightly flagged was missing in 2014. Detective Inspector Serena D'Adamo of the Met Police put the stakes plainly: "Phishing services bring misery to thousands, taking money from everyday people across the world" (The Record). The tool that stopped this one works. It just needs a procedural seatbelt, not a redesign.