Anthropic vs. Alibaba: The AI Theft Accusation Just Became a War Between Titans

Anthropic accuses Alibaba of illicitly extracting its models. Alibaba bans its own staff from touching Anthropic's tools. The abstract "distillation" debate just got two very real names

From Theory to Open Conflict

Just days ago, watchchina.ai examined "adversarial distillation" — the accusation that China builds its AI by extracting capabilities from American models — and argued that the theory, while grounded in real evidence, had become dangerously convenient for American interests. Today that abstract debate has crystallised into an open corporate war between two of the most powerful AI companies on earth, each on opposite sides of the Pacific.
The escalation runs in both directions at once. Anthropic has accused Alibaba of illicitly extracting its AI model capabilities — a direct, named allegation of exactly the kind of distillation attack that policy papers have described in the abstract. And Alibaba is set to ban its own employees from using Anthropic's coding tool, according to sources — a retaliatory move that transforms a technical accusation into a hardened commercial and national boundary. The two companies are no longer competitors sharing an ecosystem. They are adversaries drawing lines.

The Two-Front Defence

This conflict does not stand alone — it is part of a coordinated tightening by Anthropic against Chinese access that watchchina.ai has tracked in real time. As reported yesterday, Anthropic is moving to close loopholes that allow Chinese access to Claude, with engineers still finding ways to use its AI models despite stringent restrictions. The Alibaba accusation is the offensive complement to that defensive effort: on one side, Anthropic works to block Chinese users from reaching its models; on the other, it accuses China's largest cloud and AI company of having already extracted what it needed.
The timing sharpens the stakes considerably. Alibaba is not a peripheral player — its Qwen models are among the most widely adopted open-weight systems in the world, used, by one Andreessen Horowitz estimate, in the derivative products of a large share of US startups. If Anthropic's accusation is that Alibaba's competitive capabilities were built in part by extracting from Claude, then the accusation strikes at the credibility of one of China's flagship AI achievements. And Alibaba's response — banning Anthropic's tools internally rather than issuing a simple denial — is notable in itself: it is the action of a company hardening its own boundaries, not one behaving as though it has nothing to hide, though it may equally reflect a refusal to depend on a rival that has just declared commercial war.

Why Neither Side Can Fully Win This Argument

As watchchina.ai has consistently argued, intellectual honesty requires holding two truths at once — and this dispute illustrates precisely why. Distillation is a real and widely-used technique, and documented cases of Chinese IP theft exist, including the January conviction of a former Google engineer. It is entirely plausible that Alibaba, like many labs on both sides of the Pacific, has used competitor outputs to sharpen its own models. But it is equally true that Alibaba's Qwen line represents years of genuine, independent research investment that cannot be reduced to mere extraction — a purely parasitic lab does not produce one of the world's most-downloaded open model families.
The deeper significance lies in what this conflict signals about the trajectory of the entire US-China AI relationship. The era of quiet, mutually-beneficial technical exchange — where researchers on both sides built on each other's published work — is ending, replaced by explicit accusation, retaliation, and the drawing of hard corporate borders that mirror the geopolitical ones. When the world's leading Western safety lab and China's leading cloud AI company move from competition to open accusation and counter-boycott, it marks the arrival of a new phase: the bifurcation of the global AI ecosystem into two separate, mutually suspicious spheres. For the developers, startups, and enterprises that have spent years building across both — the 80% of US startups using Chinese base models among them — the ground is beginning to split beneath their feet. And China's answer to being locked out has never wavered: while Anthropic builds walls, Beijing keeps giving its models away.
Sources: The Wire China, South China Morning Post, Center for a New American Security, US-China Economic and Security Review Commission

Sources: CNBC, TechTimes, MLQ News, Memeburn, Cryptobriefing, BuildFastWithAI, Wikipedia