Enterprise & Security

China's state broadcaster goes after Anthropic's privacy policy

2 min read AI-generated

Thirteen revisions since 2023, counted by an account tied to Chinese state television. It also noticed that three data sources have become six.

Featured image for "China's state broadcaster goes after Anthropic's privacy policy"

Yuyuantantian, a social media account affiliated with Chinese state broadcaster CCTV, published a post about Anthropic on Saturday. The charge: revisions to the privacy policy have raised the risk to user data, because they allow information to be handed to US intelligence agencies without legal process whenever the company deems it necessary. Bloomberg picked the post up on Friday morning.

What the post actually says

The account did the counting: thirteen revisions of the privacy policy since 2023. Among them, provisions that move data belonging to users in Canada, Brazil, South Korea and the European Union to the United States. Between June 2024 and September 2025, it says, the number of data sources Anthropic draws on rose from three to six.

Two further points come from Anthropic’s own disclosures: the September 2025 change that made training on user data the default, and the company’s stated cooperation with US intelligence agencies over model distillation by Chinese providers.

The timing

All of this lands four days before Donald Trump and Xi Jinping meet in Washington. On Sunday, Treasury Secretary Scott Bessent and Vice Premier He Lifeng opened talks in New York with AI, tariffs and critical minerals on the agenda. Export controls and AI safety are explicitly part of it.

Yuyuantantian is known for exactly this kind of setup. The account works as a channel for positions Beijing would rather not state officially, and it has aimed at American tech companies shortly before negotiating dates more than once.

What Anthropic says

Nothing. There has been no comment so far.

The numbers hold up, the outrage was ordered in advance

Two things collide here that are worth keeping apart. A privacy policy rewritten thirteen times in three years is unremarkable on its own — every new feature, every new region, every regulator pulls a change along behind it. That training on user data has been the default since September 2025 was in Anthropic’s own announcement. Nothing in the post is new.

It is still worth reading, for a different reason. Anyone running Claude inside a company gets handed a list nobody else maintains: which clause arrived when, and which countries it touches. That the list comes from Beijing, four days before a summit, does not make it wrong. It makes it a bargaining chip in a negotiation that is about chips, not about your data.

Sources

AnthropicChinaPrivacy