Malware is hijacking Claude sessions to drain paid usage

An AI subscription is now valuable enough to farm. Anthropic is warning Claude users that infostealer malware has been stealing their active login sessions, and that the strange behavior some of them noticed, usage limits refilling and then draining while they were away, was attackers spending their quota.

How the theft works

The malware families involved are established commodity infostealers: Vidar, LummaC2, StealC, RedLine and Acreed on Windows, plus Atomic Stealer on what Anthropic describes as a small number of Macs. They copy session cookies from infected computers. Because a stolen cookie replays an already-authenticated session, two-factor authentication and single sign-on are never triggered.

Anthropic, in messages to affected users first reported by BleepingComputer on August 30, 2026, says it has no reason to believe the malware is related to Claude, installed through Claude, or connected to anything users did with Claude. The infections came from elsewhere; Claude accounts were simply among the loot.

What Anthropic is doing

The response is a clean sweep of affected accounts: forced sign-outs to kill the stolen sessions, removal of saved payment methods, and refunds for charges identified as unauthorized. The company is direct about the limits of that response, warning users that signing out stops the stolen sessions but does not remove the malware from their machines.

Why it matters

A session cookie is a password that types itself, and paid AI usage has become a resource with resale value, which puts consumer AI accounts on the same target list as banking and gaming credentials. Days earlier, a separate incident involved a fake Claude install guide hosted on a legitimate Anthropic domain delivering an infostealer. The pattern points one way: AI accounts are now standing infrastructure for their owners, and attackers treat them accordingly.