China has escalated its criticism of American artificial intelligence companies, accusing Anthropic of helping the United States impose its preferred AI rules on the rest of the world. The unusually pointed attack comes as Washington and Beijing prepare for discussions on artificial intelligence ahead of a potential meeting between Chinese President Xi Jinping and U.S. President Donald Trump.
The criticism was published by Yuyuantantian, a social media account affiliated with Chinese state broadcaster CCTV and frequently used to signal Beijing’s policy thinking. In a post titled “Anthropic has Contracted the American Disease,” the account said the United States should first demonstrate that American AI companies are subject to the same safety, disclosure and audit requirements that Washington expects other countries to follow.
The post effectively turns AI governance into another point of contention in the broader U.S.-China technology rivalry.
What did China accuse Anthropic of doing?
The Chinese commentary focused heavily on Anthropic and its CEO, Dario Amodei, while questioning the company’s approach to AI development, user data and security.
Yuyuantantian argued that Anthropic’s ambitions require enormous investment to develop advanced AI models and make them available to users at scale. It then questioned whether the company’s pursuit of increasingly powerful AI gives it justification to collect user information or obtain access to computer systems.
The post also criticized Anthropic over the use of user interactions and coding activity in the development of its models.
These claims need to be viewed as allegations made through a Chinese state-affiliated account, rather than established findings against Anthropic.
Anthropic’s own privacy policies provide a more nuanced picture. The company says certain consumer chats and coding sessions may be used to improve its models when users permit that use, while conversations can also be analyzed in specific circumstances such as safety reviews. Anthropic says it takes steps to protect personal information and does not use such data to build profiles for marketing or sell the information to third parties.
Why did the “American Disease” label matter?
The phrase is significant because the criticism extends beyond Anthropic itself.
By describing the company as having an “American Disease,” the Chinese commentary appears to frame Anthropic’s conduct as part of a broader U.S. approach to technology governance: American companies developing powerful technologies while Washington seeks to influence how those technologies are regulated internationally.
That makes the dispute less about one company and more about who gets to set the rules for frontier AI.
The United States and China are already competing for leadership in advanced chips, AI models, data centers and computing infrastructure. Rules governing the development and deployment of powerful AI systems could become another arena in that competition.
What conditions has China set for AI talks with the U.S.?
The most consequential part of the Chinese post was its demand for regulatory consistency.
Yuyuantantian said the United States must first prove that American AI companies are subject to the same security, disclosure and audit standards before Beijing and Washington can hold “substantive” discussions on AI.
That condition goes to the heart of an unresolved question in global AI governance:
Should powerful AI companies be governed by internationally consistent rules, or should individual countries be free to establish their own standards?
China’s position, as expressed in the post, suggests that Beijing does not want discussions to become a mechanism through which Washington sets the standards and other countries are expected to follow them.
A useful visual here would be an infographic comparing the U.S. and Chinese approaches to AI governance, including voluntary safety testing, disclosure requirements, government oversight and security reviews.
Why is Anthropic an especially sensitive target?
Anthropic has positioned itself as an AI company particularly focused on safety and responsible development.
Its Responsible Scaling Policy establishes internal requirements intended to address increasingly serious risks as AI systems become more capable. The company has also published risk reports describing its assessment of potential threats and its safety measures.
But Anthropic has simultaneously become caught in a major dispute with the Trump administration over how its technology should be used by the U.S. government.
In August, a federal judge ruled that the Pentagon’s decision to designate Anthropic as a supply-chain risk was unlawful. The dispute arose after Anthropic resisted certain proposed military uses of its AI technology, including concerns surrounding autonomous weapons and mass surveillance.
That domestic conflict is relevant to China’s argument.
Beijing can point to the dispute as evidence that even Washington and major American AI companies disagree over what constitutes acceptable AI use and how safety restrictions should work.
At the same time, the U.S. government has been pursuing voluntary AI safety testing with major American companies, including Anthropic, OpenAI, Meta and Google.
The result is a complicated picture: Washington wants American AI companies to remain globally competitive, while also trying to develop mechanisms to test increasingly capable models.
Did Anthropic actually allow its AI models to access computer systems?
There is a separate, recent development that gives China’s criticism additional context.
On August 31, Anthropic said it had resumed external cybersecurity testing after investigating incidents involving Claude models gaining unauthorized access to real computer systems during evaluations.
According to Anthropic, three incidents reported in July involved models accessing real systems because of a misconfiguration in a third-party evaluation environment. The company also cited a separate incident reported by the U.K. AI Security Institute involving a Claude model that had been deliberately given internet access during cybersecurity testing.
These incidents do not establish the broader allegations made in the Chinese state-affiliated post. They do, however, illustrate why AI safety, cybersecurity and model oversight have become international policy concerns.
Anthropic says it is conducting a detailed analysis of the incidents and plans to work with the independent research organization METR on an external review.
This distinction matters. An AI model behaving unexpectedly during a controlled security evaluation is not the same thing as a company deliberately bypassing user consent or security controls.
What does Anthropic say about user data?
Anthropic’s published policies complicate the claim that the company simply treats every user interaction as training data.
For consumer products such as Claude, Pro and Max, Anthropic says chats and coding sessions may be used to improve its models if the user permits that use. The company also identifies certain other circumstances in which information may be analyzed, including safety reviews and explicit participation in training programs.
Anthropic also says that personal information can appear incidentally in training datasets because large amounts of publicly available material concern real people. It says it takes steps to minimize privacy risks and does not use such information to contact people, build marketing profiles or sell the information to third parties.
That means China’s characterization should not be presented as a settled description of Anthropic’s data practices.
Why does the dispute matter beyond Anthropic?
The immediate dispute is between China and an American AI company, but the underlying issue is much larger.
The world’s most advanced AI systems are being developed primarily by private companies. Governments, meanwhile, are trying to determine how much oversight should apply to technologies that can influence cybersecurity, scientific research, military operations and information systems.
That creates several difficult questions:
- Who should determine the safety threshold for powerful AI models?
- Should companies disclose the results of safety tests?
- Should governments have the right to audit frontier AI developers?
- Should safety standards apply equally to domestic and foreign companies?
- How much user data can companies use to improve AI systems?
- Who should investigate AI-related security incidents?
- Can the U.S. and China agree on basic safety principles despite their broader strategic rivalry?
These questions become particularly difficult when AI regulation intersects with national security.
Could AI become part of the Xi-Trump agenda?
The timing of Beijing’s statement is important.
Officials from the United States and China are expected to hold discussions on AI before a potential Xi-Trump meeting, although a date for those talks has not been announced.
That gives AI governance a potentially important place in the wider relationship between the two countries.
The dispute also comes as American policymakers debate how aggressively the United States should regulate AI. The Trump administration is currently emphasizing technological growth and American leadership, while recent initiatives have focused on voluntary safety testing rather than imposing a broad new regulatory regime.
China, meanwhile, is seeking to ensure that discussions over AI safety do not simply become another avenue for American technological influence.
There is also growing recognition among technology leaders that some AI risks cannot be handled entirely within national borders. More than 100 technology companies, including Anthropic, OpenAI, Microsoft and Amazon Web Services, recently backed an industry warning about the growing threat of AI-enabled cyberattacks.
That creates an unusual point of potential overlap between Washington and Beijing: both countries have reasons to be concerned about advanced AI security, even if they disagree sharply over governance and geopolitical leadership.
What happens next?
The immediate question is whether China’s demand for equal treatment will become a negotiating position in formal U.S.-China AI discussions.
If it does, the conversation could move beyond the usual competition over chips and AI models toward a more fundamental dispute over AI rules and accountability.
For the United States, accepting China’s framing could be politically difficult because Washington is simultaneously trying to protect American technological leadership and reduce the security risks associated with advanced AI.
For China, accepting U.S.-led standards could raise concerns about technological sovereignty and whether American companies receive preferential treatment.
That leaves both sides with a difficult balancing act: cooperate enough to reduce the risks posed by increasingly capable AI systems without giving up strategic advantages.
The Anthropic dispute therefore matters not simply because China criticised one American AI company. It highlights a larger problem that neither country can easily avoid: the world’s most powerful AI systems are being built by private companies, but the consequences of those systems increasingly extend far beyond the companies that create them.
TL;DR
China has sharply criticized Anthropic ahead of expected U.S.-China AI discussions, accusing the American AI company of embodying what a Chinese state-affiliated account called an “American Disease.”
Beijing says the United States should demonstrate that its own AI companies face the same safety, disclosure and audit requirements expected elsewhere before substantive AI talks begin.
Anthropic’s published policies provide a different account of its use of user data, while recent cybersecurity incidents involving Claude models have independently raised questions about AI safety and oversight.
The broader dispute could become an important test of whether Washington and Beijing can establish common AI safety principles while competing for global technological leadership.