China investigates DeepSeek and Moonshot over an alleged data leak to Anthropic's Claude

By Zendoric · September 24, 2026. China has reportedly opened an investigation into DeepSeek and Moonshot AI, two of its most visible artificial intelligence labs, over alleged data leaks to Claude, Anthropic's model, as reported by Decrypt. That is, with the information that has reached us, practically everything that can be stated with any confidence about this episode.

The article we were able to consult —published on Yahoo News— does not give the name of the regulator that reportedly opened the inquiry, the exact nature of the alleged leak (what data, from which users, through what technical channel), the timeframe of the events or any response from DeepSeek, Moonshot or Anthropic. Given that this is a serious accusation, with regulatory and national security implications for all three parties, prudence requires treating it as such: an allegation pending verification, not a proven fact.

That said, the framework the story fits into is recognizable. Broadly, China subjects its AI labs to a data control and model approval regime considerably stricter than that of the United States, and any flow of user or training information to a foreign provider —especially a US one subject to the same export controls that already restrict Chinese access to certain hardware— is a sensitive matter for both Beijing and Washington. DeepSeek and Moonshot, the latter the creator of Kimi, are also two of the names we have cited at Zendoric as proof that China's open frontier has come very close to the Western one in capability; an investigation of this kind, if confirmed, would say nothing about that technical capability but about the regulatory friction surrounding it.

Our reading, with the caution that the scarcity of details demands: the episode fits a pattern we have been pointing out, namely that the US-China competition in AI is not fought only on benchmarks, but also over who controls the data and who audits the model. If the investigation is confirmed and its findings become known, we will learn whether there is a real leak behind it, a compliance clash between jurisdictions, or a regulatory skirmish with no technical substance. For now, with a headline as the only available source, the honest thing is to leave it there and follow the case when verifiable data emerges.