A class action against Anthropic argues that the usage limits advertised for Claude subscriptions do not hold across the whole subscription period, and that the multipliers used to describe those limits mislead buyers. Anthropic has asked the court to dismiss the case, saying every material term was available through links at the point of sign-up. What the company has not done is dispute the structure itself: its own help page sets it out, and reserves the right to restrict usage further at its discretion. That is what makes the case worth watching. The argument is not about what Claude's limits are. It is about whether linking to them counts as telling a customer what they bought.
The plaintiffs' position, as filed, is narrow. They are not claiming that the service failed, or that a limit was hit once. They are claiming that the advertised ceiling is not the ceiling that applies for the duration of the plan, and that expressing it as a multiplier implies otherwise.
Anthropic's motion to dismiss rests on disclosure rather than on the substance of the claim. The terms were linked, the argument runs, so the buyer had access to them. The help page that the plaintiffs point to is the same document Anthropic points to — it confirms the structure and adds the discretionary restriction on top.
Lawyers for the plaintiffs answered that consumers have no way to check for themselves how much service an AI product actually delivers, and are therefore forced to take the advertising at its word. The same law firm filed an earlier version of the class action at the start of the summer. That repetition matters: this looks less like a reaction to one incident than a firm working a theory into a shape a court will accept.
My reading is that Anthropic's disclosure defence is the weaker half of a strong position. Weaker, because a link at checkout is thin ground for a term that reserves unilateral discretion over the thing being sold. Stronger overall, because the plaintiffs still have to show a gap between promise and delivery, and a multiplier is a slippery thing to litigate. A multiple only means something if what is being multiplied is fixed and visible to the person paying. If it is not, the customer cannot prove they were shorted, and the vendor cannot prove they were not.
What neither side reaches, in the account of the filings so far, is a number. Nobody has said how much usage was advertised, how much was delivered, or by what margin the two diverged. Absent that, this is a case about the form of a disclosure, not about the size of a shortfall — and courts resolve form questions in ways that apply to every subscription sold the same way, not just to this one.
That is the exposure. Metered AI plans across the industry are sold in multiples and governed by help-page language that reserves discretion. If linking suffices, a subscription limit becomes a statement of intent the vendor can revise at will. If it does not, vendors will have to say what one unit of the product is — and most of them have spent two years avoiding exactly that.