Microsoft has disrupted EvilTokens, an AI-powered cybercrime platform that demonstrates how artificial intelligence is transforming online fraud from manual social engineering into an increasingly automated business model.
Key Highlights
- 12,000+ inboxes compromised across 10,000+ organizations.
- AI analyzed relationships and financial opportunities, not just phishing language.
- Cybercrime-as-a-service lowere
d the expertise required for sophisticated fraud. - Identity and session security become critical defences against AI-powered attacks.
Launched in February 2026, EvilTokens was linked by Microsoft to more than 12,000 compromised email inboxes across over 10,000 organizations worldwide, with significant victim activity in the US, Canada, UK, Australia, India and France.
Its significance lies in how AI was used. Instead of merely generating convincing phishing emails, EvilTokens could analyze compromised inboxes, understand conversations, identify financial relationships and determine which employees had authority over payments.
The platform could locate vendor invoices and wire-transfer discussions, identify organizational “money movers,” map trusted relationships and recommend people to impersonate. AI effectively helped attackers determine who to target, how to approach them and where financial opportunities existed.
Attackers reportedly obtained mailbox access through device-code phishing. Victims completed authentication through Microsoft’s legitimate sign-in process but unknowingly granted criminals account access. Importantly, stolen sessions or tokens could remain usable even after password changes unless revoked.
EvilTokens then automated work that traditionally required experienced cybercriminals to manually analyze thousands of emails. AI could summarize conversations, translate content and identify promising fraud scenarios.
The economics are equally concerning. Microsoft said EvilTokens was marketed through Telegram for a $1,500 initiation fee plus a $500 recurring subscription, effectively packaging identity compromise, intelligence gathering and fraud preparation as a service.
Investigators also found evidence that portions of EvilTokens itself were “vibe coded” using AI, illustrating a dangerous multiplier effect: AI can help criminals both build attack platforms and operate them more effectively.
Microsoft and partners ultimately seized 50 websites and disabled more than 150 associated domains. UK authorities also arrested two men in connection with the alleged operation; they were subsequently released on police bail while investigations continued.
The larger warning for CISOs is that identity compromise is becoming business-context compromise. Defences must therefore extend beyond passwords to phishing-resistant authentication, session and token protection, behavioral monitoring, identity intelligence and rapid revocation. As AI gives criminals greater contextual understanding, security teams increasingly need equally intelligent and coordinated defences.
