Malware designed to steal sensitive information has been found to copy open sessions of Anthropic’s AI chatbot, Claude, in a web browser. This allows cybercriminals to replay these sessions and access user accounts without needing passwords or two-factor authentication. In late August, Anthropic alerted users via email about the risk of compromised Claude sessions. A 7 GB file containing logs from an infostealer was shared on a Telegram channel on August 2nd. These logs included 5,871 files, each from an infected computer, spread across 162 countries. The Okta Threat Intelligence team analyzed the files, which contained usernames and session tokens. According to their report, 561 unique Anthropic authentication tokens were found on 404 infected machines, and 164 of those tokens were still valid as of August 2nd. Cybercriminals often target session tokens and API keys because they can be used to bypass traditional username and password authentication. If a session token is still valid, cybercriminals can use it to access an account directly, without triggering a two-factor authentication request. Okta highlighted a specific example involving a Windows 11 computer in Israel. The user had downloaded an infected version of the game Hearts of Iron IV, which contained session data from OpenAI and ChatGPT in a third Chrome browser profile. Okta also identified 24 still-valid API keys for four AI services, including Google Gemini and OpenAI, as reported by The Hacker News. SecurityWeek reported that some victims had their Claude accounts disconnected and credit cards removed. These attacks typically occur when users unknowingly install malicious software through unofficial downloads or apps. Fake advertisements for a product called “Claude Code” have been linked to such downloads, as noted earlier this year. Anthropic has taken steps to protect affected users by disconnecting their accounts, removing their registered credit cards, and reimbursing unauthorized charges. Some users noticed the breach when their Claude usage limits were reset and quickly exhausted, even when they weren’t using the service. In its email to users, Anthropic clarified that there was no evidence the malware was related to Claude, installed through Claude, or caused by actions taken with the AI tool. Logging out of Claude stops the stolen sessions but does not remove the malware, according to BleepingComputer. Anthropic advises users to thoroughly scan their computers to remove the malware before changing their messaging account passwords. They also recommend waiting until the system is clean before re-registering any credit cards.