Credit: © Tiero via Canva.com While the CAC summoned all 7 companies namechecked by Anthropic’s report, the probe quickly zeroed in on DeepSeek and Moonshot due to the examples the report detailed. Jing Yang wrote that on Tuesday, about her own story. She and Qianer Liu reported for The Information that the Cyberspace Administration of China is investigating both companies over user data that may have reached Anthropic. Regulators have questioned staff.
The probe is running and no penalty has been decided. The trigger was an American company’s document. TNW covered Anthropic’s threat intelligence report on 10 September, in which it said seven China-based labs had been relaying customer requests to Claude. Its published version named Moonshot AI, DeepSeek, Zhipu, Xiaomi, SenseTime and MiniMax, six of the seven.
Moonshot, it said, forwarded requests silently and showed the answers as though its own Kimi model had produced them. The figures in that report were large. Anthropic counted more than 23 million exchanges from Moonshot between May and July, and almost 300,000 requests in a single ten-day window, run through 5,380 fraudulent accounts. TNW has not independently verified the new report, and no other outlet has matched it.
The complaint has flipped direction Anthropic’s grievance was about what left Claude. Chinese labs, it said, were taking its model’s output and passing it off as their own. Beijing’s grievance is the opposite. The regulator’s concern is what went the other way: Chinese users’ data landing on an American company’s servers without those users knowing.
Same alleged conduct, read from the other end. That distinction matters, because it means China is not endorsing Anthropic’s complaint. It has found its own, inside the same evidence. What pushed the probe towards these two According to The Information, one example in Anthropic’s report involved a user the company assessed as likely affiliated with the People’s Liberation Army.
That user asked Moonshot’s Kimi to analyse surveillance footage following a person across hundreds of police cameras in Chengdu, including cameras outside PLA facilities and defence-linked institutes. Moonshot allegedly passed the request and the footage to Claude, without telling the user. TNW could not locate that passage in Anthropic’s published report. It is The Information’s account of what the report contains, and this piece carries it on that basis.
Our September coverage did record a related detail from the same document. One relayed query exposed live credentials tied to a Russian defence agency. If the Chengdu account holds, it explains why the probe narrowed the way Jing Yang describes. Police camera footage and material tied to defence institutes sit in the categories China’s own cross-border data rules treat most carefully.
A lab accused of moving that abroad is a domestic problem before it is an international one. One more detail is worth noting. Anthropic said it had identified seven labs and its published report named six. The seventh has not been identified, and The Information reports that the regulator summoned all seven.
Two weeks ago, Beijing called this sort of thing unfounded On 10 September, the day Anthropic published, China rejected a US intelligence advisory on Chinese AI distillation as unfounded. The foreign ministry dismissed the accusations in public. Twelve days later its internet regulator had summoned every company Anthropic named. Those are two different documents and it is worth keeping them apart.
The US advisory, published on 9 September by the NSA, FBI and CISA, named six Chinese AI firms. Anthropic’s report, out the following day, described seven labs. The overlap is heavy but the lists are not the same, and Beijing’s public position addressed the first. The timing is difficult for both companies DeepSeek is due to brief the United Nations Security Council this week on the risks AI poses to international security.
TNW reported that on Tuesday. The company is presenting itself as a responsible voice on AI risk in New York while its own regulator examines its data handling in Beijing. Moonshot has a different problem. It is working towards a Hong Kong listing, and an open regulatory investigation is the kind of thing that turns up in a prospectus.
Neither company has commented publicly on the probe. A third named lab is already having a bad month Zhipu, which trades as Z.ai, is on Anthropic’s list too. It spent last week dealing with a separate data problem of its own, after a developer found its ZCode tool uploading encrypted snapshots of local code repositories without consent. The company apologised on Monday, open-sourced the tool and commissioned two outside assessments.
The two incidents are unconnected and nobody has linked them. What they share is the question underneath both: when a Chinese AI product moves user data somewhere the user did not expect, who is in a position to check what happened to it. The mirror image happened in July This argument has run in both directions all year. In July, Alibaba moved to ban Claude Code over what it alleged was a backdoor risk to Chinese users.
The complaint then was that an American tool was quietly taking Chinese data. The complaint now is that Chinese tools were quietly sending Chinese data to an American company. The charge is the same shape. Only the direction of travel has changed.
What happens next is not obvious Chinese regulators have plenty of options and no stated timetable. They have not said whether they will penalise either company, and the investigation is still open. The harder question is what Beijing does with the finding if it stands. Punishing two of its strongest labs, in the week one of them addresses the Security Council, would confirm an American company’s allegation at an awkward moment.
Doing nothing leaves the data question unanswered at home. Neither option is comfortable, and there is no deadline forcing a choice.
