Anthropic Calls Chinese Labs Thieves. Built Claude on Stolen Data.

In May 2025, Anthropic published what might be the most ironic document in AI history.

They accused three Chinese labs—DeepSeek, Moonshot, and Minimax—of running an industrial-scale heist. 16 million automated queries. 24,000 fake accounts. Proxy networks designed to evade detection. Minimax literally pivoted their entire operation within 24 hours of Claude 3.5 Sonnet's release to capture capabilities before anyone could stop them.

The language was unambiguous: theft, espionage, national security threat.

Here's the thing nobody's saying out loud: Anthropic built Claude on a foundation that millions of creators argue was taken without their consent or compensation.

They just took it from humans instead.

The Foundation Everyone Ignores

Let's be clear about how frontier models actually got built.

OpenAI, Anthropic, Google—the companies now positioning themselves as responsible guardians of transformative AI—trained their models on what might be the largest IP appropriation in human history.

Books. Millions of them, scraped from Library Genesis and Z-Library without permission or payment. Living authors' entire catalogues, ingested wholesale.

News articles. Decades of journalism from outlets now fighting for survival, their archives consumed to make models smarter.

Code. Every public GitHub repository. Billions of lines written by devs who thought they were contributing to open source, not training a commercial product worth hundreds of billions.

Creative work. Fiction, poetry, screenplays, lyrics. The output of human creative lives, taken without consent, credit, or compensation.

The labs called it "fair use for transformative purposes." Courts are still deciding if that holds. The New York Times is suing OpenAI. The Authors Guild filed a class action. Visual artists organised against Stability AI and Midjourney.

Thousands of creators whose work formed the foundation of the AI economy never saw a dollar.

The labs called it training data. The creators called it theft.

The Pressure Gradient Runs Both Ways

Anthropie's argument about distillation economics is actually pretty compelling. When one side has capabilities worth potentially trillions and the other can extract them for thousands, the information moves. Always.

A 1,000-to-1 return on extraction. That's the maths Minimax was running when they pointed 13 million conversations at Claude's reasoning capabilities.

Except that's the exact same argument that justified training on the internet.

The cost of generating intelligence (human creative work, decades of writing, coding, thinking) is astronomically higher than the cost of ingesting it. A novelist spends three years on a book. A model ingests it in milliseconds.

The economics are overwhelming. The information moves.

The difference Anthropic draws is legal and technical. They own the model weights. Claude's outputs are their IP. Extracting those outputs violates their terms of service.

But the weights themselves were built on a foundation the creative class never consented to contribute.

The hierarchy nobody wants to name:

Layer 1: Human creators had their life's work taken to build frontier models. The labs called this training.

Layer 2: Frontier models now have their outputs taken to build distilled models. Anthropic calls this espionage.

The distinction is real in law. In moral logic? Considerably harder to defend.

The National Security Play

Anthropie didn't frame this as a copyright violation. They framed it as a national security threat.

Foreign adversaries. The Chinese Communist Party. Military and surveillance applications.

This framing was deliberate, and it serves specific interests.

Anthropie has consistently supported export controls on AI capabilities. They want to demonstrate those controls are working, that Chinese labs' progress depends on stolen American capabilities rather than independent innovation.

The national security frame advances a policy agenda that benefits Anthropic directly: tighter restrictions on who can access frontier AI, more deference to the companies who hold them.

Look at what DeepSeek actually targeted: Claude's reasoning capability across 150,000 exchanges, generating chain-of-thought training data at scale. Their prompts asked Claude to articulate internal reasoning behind responses, manufacturing the reasoning traces needed to train a competitor.

One technique was particularly revealing: using Claude to generate censorship-safe alternatives to politically sensitive queries. Training data designed to help DeepSeek's models steer conversations away from topics the Chinese government doesn't want discussed.

That's genuinely troubling. The geopolitical dynamic is real. The surveillance state applications are real.

But the national security frame also constructs a narrative where American AI companies are responsible guardians and Chinese labs are reckless proliferators. It obscures the degree to which the underlying economic force (the staggering ROI of acquiring capabilities through extraction rather than development) applies universally.

To every lab that isn't a hyperscaler. To every startup that can't afford a billion-dollar training run. To academic researchers, government contractors, open-source projects.

To everyone.

The Open Weights Question

Anthropie's position on open-weight models is worth examining here.

They oppose them. Not always explicitly, but structurally and in practice. Their arguments centre on safety: open weights mean anyone can fine-tune away safety guardrails, deploy capabilities without oversight.

These concerns aren't entirely without merit.

But open weights also mean capabilities built substantially on the work of millions of uncredited human creators become accessible to humans who couldn't afford training runs. Open weights mean value extracted from the global creative commons gets returned to the commons in some form.

Anthropie's opposition to open weights, framed through safety, also happens to protect their commercial position. A world where powerful open models are freely available is a world where Anthropic's API pricing faces existential pressure.

The same economic gradient that makes distillation attacks inevitable.

What This Actually Means

DeepSeek R1 just dropped and it's legitimately impressive. Whether they built those capabilities through distillation, independent research, or some combination is an open question. Anthropic's disclosure doesn't prove the capabilities are derivative, just that extraction attempts happened.

But the larger pattern is clear: information wants to flow downhill. From expensive to cheap. From restricted to accessible. From proprietary to open.

The labs built their moats on appropriated human creativity. Now they're discovering those moats are harder to defend than they thought.

The irony is absolutely unhinged.

And nobody in the AI industry seems willing to say it plainly: if taking without consent is theft when Chinese labs do it to American companies, it was theft when American companies did it to human creators.

The pressure gradient runs both ways.