Anthropic is being sued over the basic promise behind Claude Max: if you pay $200 a month for twenty times Pro usage, you should not have to guess what twenty times means.
Karl Kahn did what Anthropic wanted heavy Claude users to do. He started on Claude Pro in June 2025, moved to Max 5x in January 2026, then upgraded to Max 20x in April. The price was $200 a month. The pitch was twenty times the usage of Pro. According to a federal complaint filed June 15 in the Northern District of California, Kahn says one five-hour coding session ate through about 15 percent of his weekly allowance.
That is the whole fight. Not whether AI is useful. Not whether Claude is popular. The question is whether Anthropic sold a clear multiplier while giving customers a moving ceiling they couldn't see.
The Wall Street Journal reported that the proposed class action claims Anthropic misled customers who bought its Max 5x and Max 20x plans, which cost $100 and $200 a month. The suit says Max 20x delivered closer to six to eight times Pro usage in practice, while Max 5x delivered roughly three and a half times. Anthropic declined to comment to the Journal.
The lawsuit asks to represent U.S. residents who bought Max 5x or Max 20x through Claude.com or the Claude desktop app since April 2025. It brings claims under California's Consumers Legal Remedies Act and False Advertising Law, along with negligent misrepresentation and breach of contract, and seeks damages, restitution, and an injunction. Those are legal claims, not proven facts. But the complaint lands on a real sore point for anyone paying for AI every month: these plans are sold like subscriptions and experienced like ration cards.
The cap is the product
When Anthropic launched Claude Max on April 9, 2025, The Verge reported that the company described the plan as a way for power users to get much more usage without needing an enterprise contract. That mattered because Claude had already become a work tool for coders, researchers, founders, lawyers, analysts, and anyone else willing to pay to stop waiting on limits. If you're using Claude to write code for clients or run a research workflow, the cap is not a footnote. The cap is the product.
Anthropic's problem is that it didn't sell Max as a vague access tier. It sold numbers. Five times. Twenty times. Once you put a number in front of a customer, you invite them to measure it.
Kahn's complaint says subscribers don't get a clear dashboard showing how much of their Max allowance they have used. Claude usage also runs through more than one limiter, including five-hour windows and weekly limits. TechRadar reported in March that Anthropic adjusted five-hour session limits during peak hours for free, Pro, and Max subscribers while leaving weekly limits unchanged. For a casual user, that may be annoying. For a customer paying $2,400 a year, it becomes the billable difference between a plan that works and a plan that runs out while you're still working.
Also read: AI accounting automation will cut your bookkeeping bill in half • How to Build an AI Agent for Your Business Without Writing Code • The companies that bet everything on AI are now watching their knowledge bases quietly rot
AI subscriptions are running into math
You can see why Anthropic is tightening usage. AI models are expensive to serve, and the heaviest users are no longer just chatting. They are running coding agents, long prompts, file-heavy workflows, and repeated edits that chew through tokens. Business Insider reported this month that some Claude Max subscribers are already juggling multiple AI subscriptions, watching token use closely, or testing OpenAI tools because they worry about speed, limits, and reliability.
Frankly, that does not excuse fuzzy marketing. It explains the pressure behind it. If a company sells a $200 subscription because metered API usage would scare customers away, it still has to say plainly what the customer is buying. A vague promise might survive at $20 a month. At $200 a month, people start doing the arithmetic.
This is also why the lawsuit is bigger than one unhappy Claude user in Washington, DC. OpenAI, Google, Anthropic, and the rest are training customers to treat AI as everyday work infrastructure. Once you do that, the old consumer-app habit of shifting limits quietly starts to look flimsy. A professional user does not only need access. They need predictability.
Anthropic may defeat the complaint. The company may show that its terms gave it enough flexibility, or that Kahn's usage was unusually heavy, or that different models and prompt lengths make a simple multiplier hard to apply. Those defenses matter. But they do not erase the editorial point staring back from the complaint: if you advertise twenty times more usage, your customer should not discover the definition only after Claude tells them to wait.
The AI subscription market is now leaving its generous phase. The companies want recurring revenue, investors want growth, and users want the same thing they want from any paid service: a price, a promise, and a meter they can actually read.
