Anthropic Confirms Hackers Stole Session Keys to Drain Claude User Quotas
Anthropic confirmed that attackers used stolen session keys and info-stealer malware to generate unauthorized Claude Code OAuth tokens, allowing third parties to hijack user accounts and exhaust paid usage quotas. Multiple subscribers reported rapid quota depletion even when not actively using the service or running local tasks. This incident highlights growing security vulnerabilities around local AI agent credentials and session storage, especially as developers adopt CLI tools like Claude Code. It also underscores critical gaps in cloud billing transparency, as users were unable to track itemized usage logs to identify ongoing exploitation. Anthropic responded by revoking server-side Claude Code tokens, logging out hijacked sessions, and providing refunds in some cases, though they currently lack granular usage-tracking tools for users. The company emphasized that the info-stealer malware originated from infected software or malicious ads on user devices rather than Claude itself.
## BACKGROUND
Session keys and OAuth tokens are credentials that allow applications to maintain authentication without requiring passwords for every action. Infostealer malware often targets these cached credentials stored on local devices to bypass security measures. Claude Code is Anthropic's agentic command-line tool that integrates directly with a developer's codebase and terminal.