If you have ever paid for the top tier of an AI tool and quietly wondered whether you were getting what you paid for, a federal lawsuit filed this week is essentially that suspicion turned into a legal filing. A Washington, D.C. customer named Karl Kahn sued Anthropic on Monday, June 15, in California, accusing the company of overselling its most expensive Claude subscriptions. His claim is blunt: the plans promise a certain amount of usage, deliver far less, and are built in a way that makes it nearly impossible for a normal person to check.
What the lawsuit actually claims
Anthropic sells two premium Claude plans above its standard Pro tier, which runs around $17 to $20 a month. The first is "Max 5x" at $100 a month, marketed as roughly five times the usage of Pro. The second is "Max 20x" at $200 a month, marketed as roughly twenty times. Those multipliers are the heart of the case.
Kahn upgraded to the $200 Max 20x tier in April so he could do heavy coding work. According to the suit, a single five-hour session burned through 15% of his weekly allowance, and the usage caps started cutting him off not long after he subscribed. He says the actual usage he received was "far below the advertised amount," and that the way the plans are structured leaves a buyer with almost no way to measure whether they are getting the five-times or twenty-times usage they were sold.
The complaint leans on a set of emails Anthropic allegedly sent in July 2025. Those emails spelled out, model by model, how much weekly usage each tier should expect. The argument is that the company put specific expectations in writing, then delivered something smaller. The suit asks the court to certify a class covering everyone who bought these plans since April 2024, to order refunds, and to find the marketing fraudulent.
The part that is unproven, and the part that is real
It helps to separate what is established from what is still just an accusation. Anthropic declined to comment. The claims have not been tested in court, and no class has been certified, so for now this is one customer's complaint, not a finding against the company.
It also helps to be precise about the weekly limits themselves, because those are real and not secret. Anthropic introduced them publicly in 2025 to rein in the heaviest users, the small group running the tool almost continuously. The sharper claim in this lawsuit is a step beyond "limits exist." Kahn is arguing that the plans were sold dishonestly: that the advertised multiples and the delivered usage do not match.
There is an awkward wrinkle inside Anthropic's own product line that makes the complaint sting a little more. The company's enterprise plans, the ones sold to big organizations, already ship with granular spend caps and usage analytics. Those are exactly the tools that let a buyer see their consumption in plain numbers. The consumer tiers at the center of this suit are the opaque ones. The people most able to measure what they are paying for are the corporate accounts; the individual paying $200 a month is the one left guessing.
Why the timing matters
This lands at a sensitive moment. Anthropic and its rivals are eyeing public listings, and a fraud claim about how flagship products are sold is the kind of thing that gets attention when a company is courting investors. It also arrives just days after a US government order pulled Anthropic's top models from foreign users, so the company is already absorbing one shock to how its products reach people.
None of that decides the case. A lawsuit's timing tells you about pressure, not about merit. It does tell you the questions this raises are not going to stay quiet.
For a long time, the idea that an AI company was quietly throttling your usage or slipping you a weaker model lived in tin-foil-hat territory, the kind of thing people swore was happening with no real way to prove it. I always assumed most of that was paranoia. The useful thing about a class-action suit is that it can drag that exact question into the open, where emails get subpoenaed and numbers get put on the record. If there is something to it, this is how we actually find out.
So the takeaway is less about your refund and more about your habits. If you pay for a premium tier of any AI tool, it is worth knowing roughly what your heavy days cost you in usage, because right now the consumer products give you very little to go on. The bigger payoff here is not money back. It is that a fight like this tends to force companies to show their math, and once that becomes the expectation, the next plan you buy is a little easier to trust.