Hackers steal Claude tokens from subscribers via infostealer malware

This incident highlights a significant vulnerability in AI service accounts, where token usage is not itemized, leaving users blind to unauthorized consumption. De Swardt's experience shows that even a careful user can be victimized, and the lack of detailed usage data hampers both detection and resolution.

Anthropic's response—suspending accounts, invalidating tokens, issuing refunds—indicates acknowledgment of the issue, but the company has not yet implemented tools for users to see what is consuming their tokens. This leaves users with no proactive way to protect themselves, as De Swardt pointed out.

The stakes are high for small businesses and independent consultants who rely heavily on AI agents for core operations. For them, such an attack can disrupt their entire workflow, as it did for De Swardt, who had to pause his work and eventually leave the platform. The lack of transparency and support may drive other affected users to competitors like Cursor, which offers multiple model options. Whether Anthropic addresses this gap will likely influence user trust and retention.