The result is that victims’ accounts are being accessed and drained by the attackers, consuming their allocated usage. This type of attack affects the convenience of persistent login sessions, as the stolen tokens bypass the usual re-authentication requirements. The warning highlights that the compromise originates from the user’s device being infected, rather than a breach on Anthropic’s side.

For users concerned about exposure, the primary advice is to ensure that the endpoint is clean before taking further action. Recommended steps include:

• Run a full anti-malware scan on the affected device to identify and remove the infostealer.
• After the system is verified clean, log out of all active sessions on the Claude account to invalidate any stolen session tokens.
• Change the account password and enable two-factor authentication (2FA) if not already active.

Likely indicators that a session has been compromised include unexpected usage spikes or changes to account settings that the user did not make. While the article does not identify the specific infostealer family or assign a CVE, the core lesson is that session tokens are a valuable target and that endpoint security is a critical component of protecting AI account usage.

Source: Unknown

Is your organization auditing for unusual usage spikes in AI assistant accounts as a potential early-warning sign of session hijacking?