Infostealer malware campaign hijacks Claude login sessions to siphon subscribers' token usage

TechCrunch reports that Claude subscribers are seeing unauthorized token consumption on their accounts while inactive, and that Anthropic has confirmed at least one case where a compromised session key was used to mint unauthorized Claude Code OAuth tokens. Anthropic separately emailed other affected users warning that infostealer malware is stealing Claude login sessions from victims' computers and using them to burn account usage. Reported responses include forced sign-outs, invalidated sessions, and some partial refunds (one user cited £44.49); one named consultant, Grant De Swardt, says his account was suspended for about two weeks, disrupting his business.

This exposes a real gap in how metered AI subscriptions are secured and audited: stolen credentials can be converted directly into consumed inference quota, and without itemized usage logs, victims cannot easily detect, diagnose, or prove theft versus their own usage. That matters most for developers and businesses on usage-based plans (like the $200/month Claude Max), where account suspension or billing disputes can cause direct operational and financial harm, as in the cited two-week suspension. It is also a trust and retention issue: at least one user says he switched to Cursor over it, which is a small but real signal for vendor lock-in and switching costs in the AI coding tool market.

There is no official Anthropic public statement or security advisory here; Anthropic's account comes only through forwarded emails to affected users, so the company's official position and its estimate of scale are unverified. The scale of the campaign is not quantified anywhere (unclear if this is dozens of accounts or a wider wave), and "hackers stealing tokens" somewhat overstates the mechanism, which is standard infostealer credential theft repurposed against a new type of asset (metered API/subscription usage) rather than a novel attack on Claude itself.

1. 01Whether Anthropic issues an official public advisory or status-page statement on this campaign, rather than the current secondhand user-forwarded emails
2. 02Whether Anthropic ships itemized/per-session usage tracking for Claude and Claude Code, which would be the actual fix for detection and would validate the story's core claim
3. 03Volume signal: growing Reddit/GitHub reports, TechCrunch or other outlet follow-ups with additional named victims, or evidence of scale beyond a handful of individual cases