Anthropic has identified so-called infostealer malware – software that reads credentials from infected computers – as the cause of hijacked Claude accounts. The programs copy active login sessions and use them to take over foreign accounts and exhaust their usage limits.
Anthropic automatically logs out affected users, removes stored payment methods, and refunds unauthorized charges.
As BleepingComputer reports, citing an email from the company to affected users, at least five widespread malware programs are behind the attacks on Windows computers: Vidar, LummaC2, StealC, RedLine, and Acreed.
For a smaller number of Mac users, Anthropic also identified the malware Atomic Stealer. The malware usually enters the computer through pirated software or fake apps and reads passwords, browser cookies, and credentials from other programs, not just Claude.
With the stolen session, attackers completely bypass password and two-factor authentication because the session is already logged in; affected users often only notice the abuse when their usage limit fills up and empties again without their intervention.
However, logging out does not remove the malware from the device; affected users should change their credentials and additionally clean the computer.
The case is part of a growing series of attacks on credentials related to AI tools: just in July, a compromised npm package deliberately injected infostealers into programming tools like Claude Code, and in early August a worm spread through 444 npm packages that likewise searched for credentials of AI programming tools.
Anthropic has not yet published the total number of affected Claude accounts.
It remains open whether Anthropic will introduce additional protective measures in the future, such as binding sessions to individual devices, rather than just reacting afterwards. For office users, the case primarily shows: a clean account does not protect if one's own computer is infected – the gap lies outside of Claude.
