On November 6, 2026, a federal magistrate judge in San Francisco refused to dismiss the heart of a proposed class action over Claude’s costliest plans. Magistrate Judge Alex G. Tse held that subscribers who pay monthly could reasonably read the promise of 5x or 20x more usage than Pro as a promise about the month. The case now proceeds on three claims, while four companion claims were dismissed with leave to file them again in stronger form. Nothing has been proven, no class has been certified, and the court decided only which allegations deserve a hearing. One detail in the order, though, reaches every Canadian firm that buys AI by the seat.
- What the November 6 order decided
- Two caps sit under one headline number
- The clarification lived two hyperlinks away
- Three claims proceed, and four go back for repair
- Canada judges the general impression, not the footnote
- Budget against the weekly cap, not the multiplier
- The date that governs this case
What the November 6 order decided
The case is Kahn v. Anthropic PBC, No. 3:26-cv-05763-AGT, in the United States District Court for the Northern District of California. Karl Kahn filed the original complaint on June 14, 2026, and a consolidated complaint with Solomon Smith followed on September 8, 2026. The two subscribers sued over Max 5x at US$100 a month and Max 20x at US$200 a month, the firm’s flagship individual tiers. Pro, the comparison tier named in every advertisement for those plans, costs between US$17 and US$20 a month. Anthropic renamed the two Max tiers Max and Ultra on October 30, 2026, exactly one week before this order landed. The November 6 order, available as the November 6 order on the public docket, answers the threshold question of whether the complaint states claims worth testing.
The short answer for busy readers is narrower than the headlines suggest, and more useful than the headlines suggest. The court found no deception, awarded no damages, and certified no class of subscribers on November 6, 2026. It held instead that a headline multiplier attached to a monthly subscription can plausibly promise monthly volume to a reasonable buyer. That single holding moves the dispute from the pricing page into discovery, where delivered usage becomes evidence rather than marketing.
Two caps sit under one headline number
Every paid Claude plan carries two separate limits, one inside each five-hour session and a second one across the whole week. The advertised multiplier matches the session cap, so a Max session genuinely holds five or twenty times a Pro session. The weekly cap rises by a far smaller factor, and the week is the period a monthly subscription actually measures. The subscribers’ own arithmetic, run against the Sonnet 4 model, puts weekly delivery near 3.5x for Max 5x and 6x for Max 20x. One allegation adds that a single full five-hour window on Max 20x could consume 15% of that subscriber’s weekly allotment. Anthropic’s position throughout is that the multiplier was always per session, exactly as its help pages describe the plans. Since the order, the top tier has dropped its five-hour session limit, while every plan keeps the weekly cap underneath.
The clarification lived two hyperlinks away
The order locates Anthropic’s per-session explanation on a help page sitting two hyperlinks away from the primary sales pages on claude.com. Nobody passes through that help page on the way to subscribing, and the purchase path never puts the weekly cap beside the multiplier. The sales pages themselves carry only an asterisk reading “usage limits apply,” a warning that names no session and no week. That asterisk says nothing about how the two limits interact when a subscriber works the plan hard for a living. The court allowed that the help page might eventually persuade a fact-finder that the advertising, taken whole, was not misleading. For dismissal purposes, however, a clarification placed off the sales path cannot rescue a headline pointing the other way. The reasoning treats the buried page like the small print on the back of a shampoo bottle, present but outside the bargain.
Three claims proceed, and four go back for repair
Three claims now proceed: Smith’s claim under California’s Consumers Legal Remedies Act, the negligent misrepresentation claim, and the breach of contract claim. The contract claim survives because Anthropic’s own terms leave plan details to the order process, which plausibly includes those sales pages. The court also accepted that Claude Code can plausibly count as a service under the California consumer protection statute. Four claims were dismissed with leave to amend: Kahn’s consumer claim, the unfair competition claim, the false advertising claim, and a good faith claim. Kahn’s consumer claim fell because his stated use, programming work, could be commercial, while that statute protects personal purchases only. The unfair competition and false advertising claims fell on standing, because both men remain subscribers and cannot demand changed marketing. The good faith claim failed because the terms expressly reserve Anthropic’s right to raise or lower capacity limits as demand requires.
Canada judges the general impression, not the footnote
Canadian subscribers buy these same plans from these same pages, paying the same monthly rhythm that shaped the court’s reasoning in San Francisco. Paragraph 74.01(1)(a) of the Competition Act makes a representation that is false or misleading in a material respect reviewable conduct in Canada. The Act’s general impression test weighs the impression a representation creates alongside its literal meaning, never instead of it. A claim can therefore be literally accurate and still mislead, when the omitted fact would change what a reasonable buyer expects. Competition Bureau guidance confirms the Act reaches anyone inside or outside Canada who promotes a product or service to Canadians. This magazine has tracked the same pattern at home, where Walmart’s Sparky price pledge names no country, and only Manitoba has a law behind it, and where The Canada-EU Digital Trade Agreement is concluded. No exporter can read it yet. No Canadian court has ruled on these plan pages, and the San Francisco order binds nobody here, which is precisely why it rewards reading. The lesson travels even though the judgment does not: in Canada, the footnote does not outrank the impression the headline leaves behind. Canada’s own labelling fights run on the same discipline, as Bill C-34 labels platforms. Dec 7 campaigns still need their own AI tag shows in a different arena.
Budget against the weekly cap, not the multiplier
The procurement translation is plain enough for any Canadian firm renewing AI seats this quarter under a headline multiplier. Before your next renewal, download the limit pages for every AI plan your firm pays for and file them beside the invoices. Budget the workload against the weekly cap, because the week is where heavy agent use actually runs out in practice. Ask the vendor, in writing, which period every advertised multiplier measures, and keep the answer with the signed order form. Treat the sales page as part of the contract file, since a California court has now treated it as exactly that. The habit costs an afternoon, the same discipline this magazine urged when Budget 2026’s AI chapter isn’t for your firm. Two tax lines are became the lines that mattered.
The date that governs this case
The plaintiffs may file an amended complaint by December 2, 2026, repairing the four claims the order sent back. If no amended complaint lands by that date, Anthropic must answer the current complaint by December 16, 2026. Either way, the weekly cap is now a litigated number, and vendors everywhere are reading their own plan pages tonight. Your renewal date will arrive on the vendor’s terms unless the cap, and not the multiplier, is in your file.
