Chinese authorities have opened investigations into DeepSeek and Moonshot AI, two of the country's most prominent AI startups, over data security concerns — including whether user data flowed from the companies' services to Anthropic's Claude, according to reports from The Information, Bloomberg, and other outlets.
The Information first reported the probes on September 22, and Bloomberg followed with confirmation that the review centers on data security. Subsequent reporting by MLex and Quartz described the investigation as focused on possible transfers of user data to Anthropic's Claude — a striking twist in the escalating dispute between US and Chinese AI labs over how model capabilities are obtained. The story is a notable development for anyone tracking AI policy news, because it places Beijing's regulators in the unusual position of scrutinizing their own national AI champions. For more context on this story, see our ongoing AI trends.
The Probe Follows Anthropic's Distillation Allegations
The investigations arrived days after Anthropic's September threat intelligence report accused seven China-based AI labs of running industrial-scale campaigns to extract Claude's capabilities. According to the report, operators affiliated with Alibaba's Qwen lab injected fixed prompts into millions of requests to force Claude to write out its full chain-of-thought reasoning, then converted those transcripts into training data. Anthropic said the campaign peaked at nearly 3 million exchanges per day, launched from pools of nearly 5,000 fraudulent accounts using residential proxies, disposable emails, and virtual card payments, and totaled more than 151 million observed exchanges between May and July 2026.
DeepSeek and Moonshot were among the labs implicated — Anthropic said some of the fraudulent accounts were funneling requests from DeepSeek and Xiaomi, suggesting shared proxy infrastructure served multiple organizations. Moonshot, the Beijing company behind the Kimi model family, was named alongside DeepSeek, Alibaba, and Zhipu in the allegations, which the companies have disputed or downplayed.
Chinese regulators now appear to be examining the other side of that equation: not whether Chinese labs extracted value from Claude, but whether Chinese user data flowed out to Claude in the process.
Unusual Scrutiny of National Champions
The probe puts Beijing in an awkward position. DeepSeek and Moonshot are flagship companies in China's state-backed push for AI self-sufficiency, and both have benefited from government favorability and enterprise adoption throughout 2026. An investigation into whether their services transmitted user data to a US rival touches on China's strict data-security and cross-border data transfer rules — laws designed chiefly to keep Chinese data inside Chinese infrastructure.
If the allegations are substantiated, the compliance failure would be twofold: commercial terms violated on Anthropic's side, since the proxy-routed requests would have breached Claude's terms of service and export controls, and data governance rules breached on the Chinese side. Neither outcome is officially confirmed — reporting has relied on sources familiar with the matter, and Chinese regulators have not publicly detailed the scope or status of the investigations.
What Happens Next
Details remain sparse. The reviewed companies have not announced the probes, and no penalties or formal charges have been reported. What is clear from the reporting is that regulators are examining data flows connected to Anthropic's Claude, and that the review began after Anthropic published its allegations.
Possible outcomes range from quietly closed reviews with remediation orders to public enforcement that forces changes in how Chinese AI companies handle upstream model requests. For DeepSeek, the scrutiny lands at a delicate moment: the company just crossed $1 billion in annualized revenue and is preparing a multibillion-dollar fundraise and a STAR Market listing, processes in which regulatory goodwill is not a small variable.
Why It Matters for the Global AI Race
The episode captures how entangled the two AI ecosystems have become even as governments push to separate them. US labs accuse Chinese competitors of distillation; Chinese regulators investigate whether their own companies' data crossed the divide. Both narratives point in the same direction — model providers everywhere are tightening access, monitoring usage, and treating data flows as strategic assets.
For enterprises, the practical lesson is that API provenance is becoming a compliance issue. Requests routed through intermediaries to reach a foreign model can violate the model provider's terms, the intermediary's local data laws, or both. The DeepSeek and Moonshot probes are the first high-profile example of a national regulator policing that gray market, and they are unlikely to be the last.
Stay Ahead of AIFrom Washington to Beijing, AI policy is moving fast. Get the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
