China State Media Accuses Anthropic of Expanding Global Data Access

Yuyuantantian, a social-media account affiliated with China Central Television, has cast Anthropic’s privacy practices as a global data-security risk and as part of a broader U.S. national-security project. The central point is important to keep clear: the account’s assertions are accusations about Anthropic’s policy changes, not independently verified proof of intelligence-agency access to Claude user data.17

What Yuyuantantian alleged

According to reporting on the post, Yuyuantantian said Anthropic had revised its privacy policy 13 times since first publishing it in 2023. It argued that the changes increased user-data risk in three connected ways:17

• Transfers to the United States: The account pointed to provisions concerning transfers of data from users in Canada, Brazil, South Korea and the European Union to the U.S.
• Broader data collection: It said the policy’s listed data sources grew from three to six between June 2024 and September 2025.
• Potential intelligence sharing: It claimed the policy gave Anthropic room to share information with U.S. intelligence agencies when the company considered it necessary, rather than only after a formal legal procedure.

The post therefore presents cross-border data handling as a geopolitical vulnerability rather than as a conventional corporate privacy-policy issue. Its framing is categorical: Anthropic is portrayed as an intermediary through which foreign users’ data can become accessible to U.S. security institutions.

What the reporting does—and does not—establish

The available reporting establishes that Yuyuantantian made these claims and described its review of policy revisions. It does not independently establish that Anthropic has transferred user data to U.S. intelligence agencies without legal process, nor that any intelligence service has direct operational control over Anthropic’s systems.17

That distinction matters for readers assessing risk. A policy can describe circumstances in which a company may disclose information to authorities; proving that such a disclosure occurred, who received information, and whether it complied with applicable law requires evidence beyond an allegation about policy language.

Why Anthropic became the focus

The privacy charge follows a sharp public dispute over AI safety, model access and alleged extraction of U.S. frontier-model capabilities. In its September 2026 threat-intelligence report, Anthropic said it had disrupted malicious uses of Claude from December 2025 through August 2026 across cyber operations, influence operations, surveillance, scams and fraud, biological misuse, conventional weapons development, and model distillation.15

Anthropic also alleged that Chinese competitors, including Moonshot and DeepSeek, routed live customer chats through Claude in attempts to obtain training data and reproduce its capabilities. Reuters reported those allegations, which are Anthropic’s findings rather than adjudicated conclusions about responsibility by the Chinese government.1

In the same reporting cycle, Anthropic described activity by China-based actors involving military-related analysis, including work connected to electronic warfare and air-defense suppression. Again, these are company-attributed findings, not publicly adjudicated determinations.2

A contest over the meaning of AI safety

Seen together, the two sides are emphasizing different threats:

• Anthropic’s framing centers on misuse of powerful models, cyber and weapons-related applications, and unauthorized distillation of model capabilities.115
• Yuyuantantian’s framing centers on the privacy implications of U.S.-based AI services, cross-border data transfers and the possibility that security justifications could enable state access to user information.17

This is why the dispute is larger than a disagreement over a privacy notice. It is a contest over who gets to define AI risk. One narrative treats frontier models as strategic technologies that must be guarded against extraction and misuse. The other treats the operators of those models as potential vectors for surveillance and foreign control over data.

What users should take from the dispute

For organizations and individuals evaluating AI services, the practical lesson is to separate rhetoric from verifiable policy and operational controls. Check which product terms apply, where data may be processed, what retention and training choices are available, and under what stated conditions a provider may disclose information to authorities.

The public claims in this episode do not resolve those questions on their own. They do show that privacy terms, model-security controls and data-location commitments have become central issues in the geopolitical competition around AI. As that competition intensifies, clear documentation and independently verifiable safeguards will matter more than political narratives from either side.