Anthropic has deliberately cut off a meaningful slice of its own paying customers — accounts representing between 5% and 7% of its revenue in the first nine months of 2026 — because they were Chinese AI labs quietly mining Claude to train competing models. The disclosure came at an investor meeting in San Francisco last week, where Anthropic executives explained a slowdown of their own making, Reuters reported on October 22.
Weeks before a possible November IPO that could be the largest in history, Anthropic is choosing to walk away from revenue rather than let rivals keep extracting its technology. The company told investors the crackdown has effectively stopped Chinese developers from distilling its models — and gave itself a tidy explanation for a month-to-month growth deceleration that was already making investors nervous.
What Anthropic told investors
Inside the investor meeting, Anthropic’s executives framed the move as a deliberate trade: the blocked accounts were not typical customers but operations using fraudulent accounts to run huge volumes of queries through Claude and feed the outputs into training runs for rival models. Investors at the meeting told Reuters the affected accounts made up 5% to 7% of Anthropic’s revenue over the first nine months of the year.
The disclosure matters because the numbers behind it are extraordinary. Anthropic brought in $4.7 billion in a single quarter by this past March, then climbed to $11.6 billion in the second quarter and $16.8 billion in the third. Growth at that scale rarely keeps compounding forever, and investors have been watching the deceleration closely.
Reuters also reported that some investors are now discussing a valuation near $1.5 trillion for the IPO, down from the roughly $2 trillion once expected. Anthropic has not publicly confirmed a filing date, though reporting has said a public S-1 could arrive as early as this week.
The distillation campaigns behind the crackdown
Distillation is the practice of training a smaller “student” model on the outputs of a larger “teacher” model, absorbing its capabilities secondhand. In February, Anthropic said Chinese labs including DeepSeek, Moonshot and MiniMax had generated more than 16 million interactions with Claude through roughly 24,000 fake accounts, all in violation of its terms of service.
Anthropic has kept returning to the subject. In an October 12 research post, the company said a model distilled from Claude’s full reasoning transcripts scored 80.4% on SWE-bench Verified, versus 71.8% when distilled only from summarized reasoning — evidence, it argued, that the reasoning traces themselves, not just final answers, had become the real prize for distillers.
That post named a wider set of beneficiaries of extracted data, including Zhipu’s GLM-5.3 and Moonshot’s Kimi, and said tens of thousands of accounts tied to distillation campaigns had been banned, often within two hours of a campaign being detected. Anthropic described the practice as more than a contract violation — a loss of intellectual property with national security implications.
New defenses and the claimed 90–99% reduction
Anthropic told investors it has rolled out a bundle of defenses this month: access restrictions, proxy blocking and faster account bans. The company said the measures have cut the amount of Claude reasoning data being siphoned off by roughly 90% to 99%, effectively stopping Chinese developers from distilling its models.
Those figures are Anthropic’s own claims, and independent replication is not available. What is verifiable is the scale of the enforcement: tens of thousands of accounts banned, some within hours of a campaign being detected. The company has also moved on the policy front, updating its terms of service to block access for companies majority-owned or controlled by Chinese entities.
Why It Matters
Anthropic is heading into the biggest AI IPO ever with an unusual pitch: part of its recent growth came from rivals paying it to copy it, and it is now choosing to sacrifice that revenue to protect its lead. For publishers and developers watching the AI race, the message is that the frontier labs see model theft not just as a security problem but as a business-model problem worth billions.
The crackdown also sharpens the US–China technology split. Chinese labs face export controls on the most advanced chips and now a direct ban from Claude’s API. Anthropic is betting that cutting off access will slow their progress. Whether it buys a lasting lead or just a temporary edge is the question investors will price into the listing.
Background
Anthropic, founded in 2021 by former OpenAI researchers, has spent the year positioning itself as the AI safety company — pitching a draft federal AI safety bill, committing 5% of its shares to a public benefit fund and granting METR researchers employee-level access to assess Claude’s risks. The distillation fight sits inside that same posture: a company arguing its models are too important to be copied wholesale, and willing to lose paying customers to prove it.
Source: Reuters (reporting from the investor meeting; The Information independently reported the same disclosure)
0 Comments