28.8M Claude Pulls Ignite U.S.–China Fight

Anthropic says Alibaba-linked operators ran a massive Claude extraction campaign that turned an AI rivalry into a national security fight.

Quick Take

Anthropic accused operators linked to Alibaba’s Qwen lab of using nearly 25,000 fraudulent accounts.
The company said those accounts made 28.8 million Claude exchanges between April 22 and June 5, 2026.^[1]^[9]
The targeted activity focused on software engineering and agentic reasoning, which are core Claude strengths.^[1]^[9]
Anthropic says this was illicit “adversarial distillation,” not normal use of its service.^[1]

What Anthropic Is Claiming

Anthropic has publicly accused operators linked to Alibaba Group and its Qwen AI lab of running what it calls a major distillation campaign. The company says the effort used fraudulent accounts and proxy-style access to pull Claude outputs at scale, then reuse them to improve another model. Anthropic told senators and White House officials that the campaign was the largest of its kind it had seen from a Chinese firm.^[1]^[9]^[10]

The numbers are the heart of the charge. Anthropic says the activity covered 28.8 million exchanges with Claude and roughly 25,000 fake accounts over a little more than six weeks.^[1]^[9]^[10] It also says the prompts centered on software engineering and agentic reasoning, which are the kinds of tasks that make frontier models valuable to developers and businesses. That makes the allegation more than a simple terms-of-service dispute.

Why Anthropic Says This Matters

Anthropic has argued that distillation can be legitimate when a company compresses its own model. But it says the same method becomes a serious problem when rivals use outputs from a stronger system without permission. In earlier research, Anthropic said it had already detected industrial-scale distillation attacks by three Chinese laboratories and described those campaigns as a national security risk because they can strip away safety controls built into American models.^[20]^[22]

That framing will resonate with readers who are tired of seeing U.S. technology undercut, copied, or exploited by foreign competitors. Anthropic says the concern is not just theft of ideas. It says model extraction can give outside labs a faster path to advanced systems without paying the full cost of research, testing, and safety work. In other words, the company is arguing that the process rewards cheating while weakening the guardrails that protect users.

What Is Confirmed, and What Is Not

What is confirmed is that Anthropic itself made the accusation and released a detailed account of prior distillation cases involving other Chinese labs.^[1]^[20] What is not confirmed in the public record is Alibaba’s side of the story. The available reports do not show a formal public rebuttal, and there is no public court filing or government charge against Alibaba tied to this claim. That leaves the accusation serious, but still one-sided in the public arena.^[2]^[9]^[11]

Anthropic accused Alibaba-linked operators of the largest distillation attack yet…

~25,000 fake accounts + 28.8 million interactions with Claude (April–June 2026) to steal advanced reasoning, coding & agentic capabilities.

This hits the core global issue.

Why does the…

— SunDeep Mehra (@SundeepMehra7) June 25, 2026

Anthropic also said its safety systems now watch for distillation behavior and route suspicious traffic away from Claude’s lower-tier model. The company says its newer model line uses conservative safety classifiers and that most sessions do not trigger fallback behavior. That suggests Anthropic believes it can spot abuse patterns early, but it also raises the obvious question: if the company can detect the abuse, it should be able to document the evidence clearly and withstand scrutiny.

The Bigger Fight Over Frontier AI

This dispute fits a larger pattern in the AI race between the United States and China. American firms are increasingly accusing Chinese labs of using mass querying, fake accounts, and proxy access to mine frontier models for training data.^[17]^[20]^[22] For conservatives who already worry about weak borders, loose oversight, and U.S. companies being exploited by foreign actors, the parallel is hard to miss. The battle is now about who controls the most powerful tools of the next economy.

Alibaba has not, in the public record provided here, issued a detailed technical denial that answers Anthropic’s specific claims. That silence may be strategic, but it also leaves room for skepticism on both sides. Anthropic’s accusation is specific and high stakes, yet the public evidence still depends on its own account. Until Alibaba responds or outside investigators verify the logs, this remains a major allegation, not a proven case.