Anthropic signs out Claude users after infostealer attacks

Anthropic warned users that infostealer malware had been used to take active Claude login sessions.[1] With those sessions, attackers could enter accounts and consume users’ available usage limits.[1] The company said the attacks were launched from compromised computers and involved common infostealer malware, not a vulnerability in Anthropic’s system.[1]

Anthropic said it was revoking compromised sessions and signing affected users out.[1] It also deleted affected users’ saved payment information and said it would refund charges that its investigation determined were unauthorized.[1] The company’s response, reported as occurring on September 2, 2026, included both the sign-outs and payment-information deletion.[2]

Anthropic said unusual activity around usage limits may indicate unauthorized account access.[1] One example it gave was a limit resetting and then being exhausted rapidly.[1] On August 4, Claude Max 20x user Grant De Swardt, an independent AI consultant in East Sussex, U.K., noticed unexpected account activity; he said his token usage was climbing on a day he had not been working.[4] That account of unexpected usage illustrates the kind of activity users may want to check, but the evidence does not say that every unusual usage pattern means an account was compromised.[1][4]

Anthropic said account-level actions cannot remove malware already present on an infected computer.[1] It urged users to remove malware and change relevant credentials, and advised caution when downloading software, especially from unofficial sources.[1] The company identified the Windows infostealers Vidar, LummaC2, StealC, RedLine and Acreed, and said it also found Atomic Stealer, or AMOS, on some Macs.[1] The attacks mainly targeted Windows PCs and some Macs; phones were unaffected.[3] A user’s account actions should therefore be paired with checking the device itself.[1]