If you run Claude Code hard, you have felt the walls close in. A session that used to carry a whole afternoon now eats a fifth of your week. The usage bar tops out before you have shipped anything. And every time the community said so out loud, the reply was some flavor of “limits haven’t changed, this is working as intended, about 5% of users are affected.” On June 14, that feeling stopped being a vibe and got a docket number.

What the Complaint Says

Kahn v. Anthropic PBC, filed June 14, 2026 in the Northern District of California, is a putative class action covering everyone who bought Max 5x ($100/mo) or Max 20x ($200/mo) since they launched in April 2025. The core allegation is blunt: the plans deliver usage “far below the advertised amount,” and Anthropic “entices consumers to pay the $200 monthly Max 20x subscription by falsely claiming it offers a 50% savings.”

The plaintiff, a D.C. coder named Karl Kahn, says a single five-hour session burned 15% of his weekly allowance, and that he kept buying extra usage after hitting caps he was never clearly shown. The legal theories are California’s Consumers Legal Remedies Act, its False Advertising Law, negligent misrepresentation, and breach of contract. Anthropic declined to comment.

You Weren’t Imagining It

The reason this lands is that the timeline is a paper trail, and it matches what every heavy user swears they felt:

  • September 2025: Sonnet 4.5 ships, and the weights used to meter your quota tighten silently. Usable hours drop from 40-50 a week to 6-8. No announcement.
  • January 2026: the holiday “double limits” bonus expires and developers report a roughly 60% reduction, some hitting 100% usage “within 10-15 minutes.” Anthropic’s framing: complaints are “largely a response to the resumption of normal limits.”
  • March 2026: peak-hour throttling tightens, a bug makes the model discard its own reasoning mid-session and drain tokens faster, and a Claude Code release burns about 40% more tokens than the prior one. A Max 20x user watches usage jump 21% to 100% on one prompt.

Three distinct episodes, three distinct “it’s in your head” responses. It was never in your head. It was in the changelog you weren’t given.

”It’s Fine” Was the Posture

What stings isn’t only the throttling. It is the ten months of insisting it wasn’t happening. The communications ran from denial-by-omission to deflection to a 5%-then-7%-affected statistic to the reliable refrain that “weekly limits remain unchanged” - technically true while the five-hour window quietly emptied faster. The admission, when it finally came, was almost a relief to read:

This isn’t the experience users should expect from Claude Code. Demand for Claude has grown at an unprecedented rate, and our infrastructure has been stretched to meet it, particularly at peak hours.

— Anthropic engineering postmortem, April 2026

There it is. Not “you imagined it.” Not “5% of users.” Stretched infrastructure, at peak hours, for ten months, while the public line was that nothing was wrong. The trust erosion was never about the existence of limits. It was about being told the limits weren’t moving while you watched them move.

The Meter You Couldn’t See

The real grievance is the hidden gauge

Limits are defensible. Compute is finite and the heaviest users genuinely strain it. What is not defensible is running a meter the customer is not allowed to read. Anthropic never disclosed the actual quota structure, so users reverse-engineered an eight-bucket system and shipped ClaudeMeter to see what their own subscription was doing. The most-upvoted “fix” on the forums was teaching Claude to answer like a caveman to spend 75% fewer tokens. When your customers are building telemetry you withhold and cosplaying as cavemen to afford the plan they already pay for, the product is the opacity.

This is also why the lawsuit is more dangerous than the usual SaaS grumble. Vague promises (“unlimited,” “blazing fast”) are hard to litigate. But “5x” and “20x” are numbers. They are comparative, quantifiable, and testable against delivered usage. If discovery shows the multiplier never matched the meter, the false-advertising theory has something most consumer suits never get: arithmetic.

What This Isn’t

I want the cynicism bounded by the facts:

  • One suit, not a movement. Kahn is a single putative class action, not yet certified. No FTC, no state AG, no EU. It may settle quietly or die on the pleadings, and the breach-of-contract claim is weak because the terms of service reserve the right to change limits.
  • The scarcity was real, not invented. This wasn’t pure greed. Demand genuinely outran compute, and when Anthropic landed the SpaceX Colossus deal in May 2026, it raised limits and dropped peak throttling almost overnight - retroactively confirming the constraint had been real all along.
  • They did, eventually, admit it. The April postmortem and the May limit increases are real accountability, late as they were. The sin was the silence, not the ceiling.

The Lawsuit Already Won Something

Kahn v. Anthropic might not survive a motion to dismiss. The class might never get certified. But it has already done the one thing ten months of forum threads could not: it put “show us the meter” in front of a judge. The undisclosed buckets, the silent reweightings, the gap between “20x” and what 20x actually bought - those are now discovery requests, not Reddit speculation.

You felt the walls move. Now Anthropic has to explain, under oath, whether they did. That was never in your head.