Anthropic's worst week: token theft and a subscriber lawsuit

Anthropic has had a rough few weeks. In the same stretch of time, the company has had to acknowledge that a bad actor was covertly stealing tokens from paying Claude subscribers, and face an expanded class action lawsuit alleging that its Max plan pricing was deliberately obscured. Neither story is flattering on its own. Together, they paint a picture of a company struggling to maintain the confidence of the very power users it has publicly said are central to its business.

What makes this particularly uncomfortable for Anthropic is that both crises involve its highest-paying customers. The Claude Max plan, which costs either $100 or $200 per month, is the product at the centre of both the theft complaints and the lawsuit. These are not casual free-tier users. These are people and businesses that have committed serious money to Anthropic's platform, and in the same week they learned their tokens could be stolen and their subscription terms may have been misrepresented.

Infostealer malware used to drain Claude Max token allowances

The theft story broke after Grant de Swardt, an independent AI consultant in East Sussex, noticed his Claude Max 20x account consuming tokens on days when he had done no work and had disabled all connected services. He contacted Anthropic and asked for an itemised usage log. Anthropic could not provide one, which is itself a significant gap in account transparency, but the company did investigate. It found that a compromised Claude session key had been used to mint unauthorised Claude Code OAuth tokens, allowing a third party to consume his allowance without his knowledge. Anthropic suspended his account, invalidated his sessions, and refunded him £44.49 of a $200 per month subscription. When de Swardt posted about this on Reddit, dozens of other Claude users reported similar experiences, including accounts that had been auto-upgraded without consent and usage that jumped from zero to 100 percent overnight. Anthropic confirmed in emails to affected users that infostealer malware was being used to harvest Claude login sessions from computers, which were then used to access accounts and burn through token allowances. The company's response, to its credit, included proactive warnings to affected users, but the absence of itemised usage data means this kind of theft could run for weeks undetected by anyone who is not watching their percentage gauge closely.

Class action lawsuit targets Claude Max subscription advertising

Separately, a group of Claude subscribers filed an expanded class action complaint alleging that Anthropic's marketing of the Max plan is deceptive. The plan advertises "5x" or "20x" the usage limits of the $20 per month Pro plan. The lawsuit, brought by attorneys Monica Vaca and Kati Daffan, both formerly of the FTC under Lina Khan, argues that those multipliers are only meaningful within five-hour rolling windows that are also subject to a weekly cap. Reaching that full definition requires clicking through two separate hyperlinks from the pricing page, and then visiting the Pro plan page separately to understand what "session" actually means. Anthropic's defence is that the information is technically available if a customer follows the links, describing it as "the digital equivalent of flipping a product over to read the back label." The complaint was originally filed in July, withdrawn, and refiled as a class action. Anthropic did not introduce the contested weekly limits until August, several months after launching the Max plan in April 2025.

Why power users are now Anthropic's biggest liability

Anthropic has explicitly positioned power users as a priority, even making decisions that cut off third-party integrations in order to protect that segment. But the events of this week reveal a structural problem with that strategy. Power users are not just higher-revenue customers. They are also more likely to notice when something is wrong, more likely to compare notes on Reddit and GitHub, and more likely to have the technical literacy to document and escalate. De Swardt's Reddit post generated 80 replies within hours. The lawsuit attorneys say they heard repeatedly from subscribers who upgraded mid-project after hitting limits, then discovered the expanded capacity was not what the marketing implied. These are exactly the users Anthropic needs for long-term retention, and they are now the loudest critics.

What this means for anyone evaluating Claude subscriptions

If you are currently on Claude Max, or considering it, there are two practical takeaways. First, monitor your token usage actively and frequently, because Anthropic does not currently offer itemised usage breakdowns on request, which means unauthorised consumption is hard to detect until significant damage is done. If you use Claude Code or have connected third-party services, review your active OAuth tokens and rotate your session credentials. Second, before paying for a Max tier subscription, read the full session and weekly limit definitions, not just the headline multiplier on the pricing page. The lawsuit may or may not succeed, but the underlying confusion it describes is real and well-documented by users. Anthropic has said it is hearing feedback on price and value, as evidenced by comments in its Fable 5.1 release notes, but structural transparency on both usage security and subscription terms needs to improve before that message will land.