OpenAI, Anthropic warn of Chinese fake-account campaigns copying their AI

OpenAI and Anthropic have warned that Chinese AI labs are using tens of thousands of fake accounts to extract capabilities from their flagship models, escalating tensions in the U.S.-China technology rivalry at a moment when American AI is simultaneously flowing to Chinese firms through legal loopholes.

Distillation Campaigns Target U.S. AI Leaders

The warnings from the two leading AI labs center on so-called "distillation attacks," in which operators use a more powerful AI model's outputs to train a less capable one. Anthropic in June accused Alibaba ↗ 1.07% of conducting "the largest operation to unlawfully extract Claude's capabilities," telling Senators Tim Scott and Elizabeth Warren that individuals linked to Alibaba used roughly 25,000 fraudulent accounts and generated more than 28.8 million exchanges with Claude between April 22 and June 5, 2026. OpenAI has raised similar concerns about coordinated extraction efforts targeting its own systems.reuters+3

The companies, along with Alphabet's ↘ 0.34% Google, began collaborating earlier this year through the Frontier Model Forum — an industry nonprofit they co-founded with Microsoft ↗ 0.19% — to detect and counter adversarial distillation attempts that violate their terms of service.mercurynews+1

U.S. AI Models Still Reaching Blacklisted Chinese Firms

In a parallel development first reported by the Financial Times on July 9, OpenAI and Google confirmed they supply advanced AI services to Singapore-registered subsidiaries of Alibaba, Baidu ↗ 0.02%, and Tencent — all of which appear on the Pentagon's "1260H" list of firms with alleged ties to the Chinese military. Under current U.S. export controls, mainland China is restricted but Singapore is not, creating a gap that allows the sales to proceed legally.ft+2

Current rules block direct access to a handful of frontier models, including OpenAI's GPT-5.6 and Anthropic's Mythos and Fable, but stop short of a blanket ban on Chinese-headquartered firms purchasing cutting-edge AI software.thenextweb

A Widening Contradiction

The juxtaposition highlights a growing contradiction in Washington's approach to AI competition: U.S. labs are simultaneously sounding alarms about illicit Chinese model copying while legally selling advanced AI to the same conglomerates accused of those extraction campaigns. Alibaba has filed a legal challenge to its Pentagon designation, further complicating any potential enforcement action. OpenAI has said it prohibits access from within China but allows "some companies" with Chinese ownership to use its AI services from permitted jurisdictions.yahoo+1