AI Tech / news
Anthropic Warns Claude Users After Subscriber Reports Token Theft
An AI consultant in the U.K. discovered his Claude Max account burning tokens while he was idle, prompting an investigation and a warning from Anthropic about hackers.
Grant De Swardt, an independent AI consultant based in East Sussex, U.K., first noticed the anomaly on August 4. Despite not working that day, his Claude Max 20x subscription showed climbing token usage. The next day, he disabled all integrated tools and avoided using the platform, yet consumption still rose.
During what De Swardt described as the clearest controlled period, token usage increased from 45% to 55% while he performed no work, with scheduled Cowork tasks paused or completed, Dispatch/cloud execution disabled, and no active local Claude Code task. He contacted Anthropic for an itemized breakdown, but the company did not provide one.
Anthropic acknowledged the issue, suspending his paid account, invalidating sessions and server-side Claude Code tokens, and issuing a partial refund of £44.49 for the remaining time on his $200-per-month subscription. The disruption affected his business, as he relies on AI agents for daily administration, website design, coding, and setting up similar systems for small and mid-sized companies.
The incident has led Anthropic to warn users about hackers targeting Claude tokens. De Swardt's experience highlights the potential vulnerability of subscription-based AI accounts to unauthorized usage.