Anthropic: Up to 7% of Revenue ($2.33 Billion) Came From Customers Who Were Later Banned
Anthropic admitted in its IPO prospectus that 5–7% of its revenue for the first nine months of 2026 came from customers it later deemed violators. That is $1.66–2.33 billion out of $33.24 billion. The figure covers payment for services delivered before the ban. The document does not say how much money the company kept without refunding it. Under the terms, though, a subscription is not refunded if you are banned for a violation, so pay month to month and keep your correspondence.
AI-processed from Habr AI; edited by Hamidun News
Picture this: a developer pays $200 for a Claude Max plan, launches Claude Code in the terminal, and 31 minutes later loses access over "suspicious signals." The appeal is rejected with a form letter, and the money is not refunded. Stories like this have circulated in chats and on GitHub for a while, but now they have a financial context.
Ahead of its Nasdaq listing under the ticker ANT, Anthropic filed an S-1 prospectus, and an amended S-1/A, with the U.S. Securities and Exchange Commission.
The filing says 5% to 7% of revenue for the first nine months of 2026 came from customers later found to have violated the rules. On total revenue of $33.24 billion, that is $1.
66 billion to $2.33 billion. If you pay for Pro, Max or the API, this concerns your money and your risk directly.
Figures in such documents can't be fudged: deception carries liability, including criminal liability.
What the prospectus actually says
An important caveat: this is revenue for services delivered before access was restricted. It is not the amount the company kept after the bans. The prospectus does not say how much money was not returned to banned customers. There is also no breakdown between enterprise customers and ordinary Pro and Max subscriptions. For scale: the upper bound, $2.33 billion, is roughly half of the company's entire 2025 revenue ($4.585 billion). Revenue in 2024 was $386 million, and over the first nine months of 2026 it grew 1206% year over year.
The document treats distillation, where other labs train their models on Claude's outputs, as a separate risk factor. The company does not tie it to the full 5–7%. Anthropic also warned investors that identifying and restricting violators could reduce future revenue and growth rates. It did not say by how much or when.
The scale of the bans is large. According to Anthropic's own report, 11.4 million accounts were banned for all types of violations in January–June 2026, an average of about 63 thousand per day. According to an analysis on Habr, about 398 thousand appeals were filed and roughly 42 thousand decisions were reversed, or about 10.5% of appeals.
The legal fork: suspicion is enough, a refund is not guaranteed
The terms for individuals say the company may suspend or terminate access at any time without notice if it believes the terms have been violated. If access is closed for that reason, the subscription is not refundable. Prepaid API credits are likewise declared non-refundable. If Anthropic cancels a subscription bought on its website for any other reason, the money is refunded pro rata for the remaining term. Purchases made through app stores follow the store's rules, and local law may grant additional rights.
The acceptable use policy allows Anthropic to warn, restrict or disable access when it suspects a violation. Suspicion alone does not make a decision wrong. But judging by complaints, the reason is often not explained, and appeals get form replies.
Examples from public complaints, which cannot be independently verified:
- An engineer on GitHub paid $200 for Max 20x, was cut off 31 minutes later, and was refused a refund.
- A developer paid S$186.59 by Visa card and was banned in under a day.
- A user paid $216 for Max and lost the account on the day of purchase, with no refund.
- A Russian author on Habr got access back three hours after appealing and recovered $78.97 for the unused time.
These are individual stories, not statistics: you can't calculate how often each outcome occurs from them.
What isn't proven, and what to do
You can't claim that Anthropic bans people to save money. True, in the third quarter of 2026 the company posted an operating loss of $567 million, and its commitments for compute and data centers total about $900.9 billion, of which about $70.7 billion falls in the next 12 months. A banned paying customer no longer loads the servers, but the costs tied to that customer are not zero, and the prospectus does not disclose the economics of an individual ban.
What to do in practice:
- Pay month to month and don't buy large API packages in advance: your maximum loss is limited to one month.
- Keep your card's country, your region and the service terms consistent. How the checks work exactly is unknown from the outside, so this is no guarantee.
- Save emails, receipts and support correspondence: you will need them for an appeal and for a dispute with your bank.
- Don't tie your workflow to a single provider, and keep a backup model for critical tasks.
For a product manager or founder, the takeaway is simple: refund terms are part of a subscription's real cost, alongside its price.
- Nov 19, 2026Giant US Data Centers Use 82 TWh a Year — More Than Each of 148 Countries
- Nov 19, 2026A Lawyer Taught an AI Agent to Make Surgical Edits in Word with Track Changes
- Nov 19, 2026RSAC 2026 in Hindsight: AI Assistants Invented 19.7% of Recommended Packages
- Nov 19, 2026FT: To Pay Off $1 Trillion in Contracts, Anthropic Needs $602 Billion in Annual Revenue
Key Questions Answered
What the prospectus actually says?
An important caveat: this is revenue for services delivered before access was restricted. It is not the amount the company kept after the bans. The prospectus does not say how much money was not returned to banned customers. There is also no breakdown between enterprise customers and ordinary Pro and Max subscriptions. For scale: the upper bound, $2.33 billion, is roughly half of the company's entire 2025 revenue ($4.585 billion).
What are the key details on The legal fork: suspicion is enough, a refund is not guaranteed?
The terms for individuals say the company may suspend or terminate access at any time without notice if it believes the terms have been violated. If access is closed for that reason, the subscription is not refundable. Prepaid API credits are likewise declared non-refundable. If Anthropic cancels a subscription bought on its website for any other reason, the money is refunded pro rata for the remaining term. Purchases made through app stores follow the store's rules, and local law may grant additional rights.
What isn't proven, and what to do?
You can't claim that Anthropic bans people to save money. True, in the third quarter of 2026 the company posted an operating loss of $567 million, and its commitments for compute and data centers total about $900.9 billion, of which about $70.7 billion falls in the next 12 months. A banned paying customer no longer loads the servers, but the costs tied to that customer are not zero, and the prospectus does not disclose the economics of an individual ban.
Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews
Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.