Chinese authorities are investigating two of the country's most prominent AI companies, DeepSeek and Moonshot AI, over suspected leaks of data to Anthropic, according to a report from The Information. The probe signals a significant escalation in how Beijing is responding to allegations that its domestic AI firms have been entangled with American competitors, whether through deliberate routing, distillation, or unauthorized data transfers.
What the Investigation Covers
The exact scope of the Chinese government's inquiry has not been made fully public, but sources cited by The Information indicate regulators are examining whether user data or proprietary model information found its way to Anthropic, the San Francisco-based AI safety company behind the Claude series of models. Both DeepSeek and Moonshot have denied wrongdoing in previous public statements, though neither company has addressed the latest probe in detail.
Key Facts
- Chinese regulators have opened formal probes into DeepSeek and Moonshot AI.
- The investigation centers on potential data leaks to Anthropic.
- Both firms have previously denied routing user traffic to Claude or using its outputs improperly.
- The probe comes amid broader US government scrutiny of Chinese AI companies and their alleged use of American AI models.
- Anthropic has not commented publicly on the Chinese investigation.
The timing is notable. US officials have spent recent months raising alarms about Chinese AI companies allegedly exploiting American models. A Trump administration official previously accused Moonshot AI of stealing from Anthropic, framing the issue as a matter of intellectual property and national security. Now Beijing appears to be conducting its own version of due diligence, though its motivations may differ considerably from Washington's.
The investigation reflects a pattern in which both governments are scrutinizing the same companies, but for potentially opposite reasons.The Information
A Complicated Web of Allegations
The backstory here is layered. Earlier reporting established that Anthropic had been looking into claims that Moonshot and DeepSeek were rerouting user requests to Claude rather than serving responses from their own models. Separately, Anthropic detailed what it described as distillation attacks, a technique where outputs from a powerful model like Claude are used to train a competing model. Those findings put both Chinese firms in an uncomfortable position with American regulators and the public.
China's investigation flips the framing. Rather than asking whether Moonshot or DeepSeek were copying from Anthropic, Beijing is asking whether data was flowing outward to Anthropic without authorization. That could mean user queries, behavioral data, or other proprietary information leaving Chinese systems and ending up with an American company. If substantiated, such a finding would carry serious implications under China's data sovereignty laws, which treat cross-border data flows as a sensitive national security matter.
For Anthropic, the situation presents a reputational and diplomatic puzzle. The company has positioned itself as a safety-focused lab committed to responsible AI development. Being named, even indirectly, in a Chinese government probe is an unwelcome complication, regardless of whether any wrongdoing occurred on its part. Those following the broader story of how Claude has been drawn into the US-China AI rivalry will recognize this as another chapter in an increasingly complex dispute.
What Comes Next
It remains unclear what penalties, if any, Beijing might impose on DeepSeek or Moonshot if the investigation finds violations. China has used its data protection and cybersecurity frameworks aggressively in the past, including against domestic tech giants. For AI companies still building their reputations and user bases, a government sanction could be damaging even if the underlying facts are disputed.
The situation also puts pressure on the broader AI industry to clarify how data moves between systems, especially when models from different countries interact, either directly or through API calls. Transparency around those data flows has been inconsistent across the industry. As governments on both sides of the Pacific treat AI infrastructure as a strategic asset, the days of ambiguity may be running short. For readers tracking how these developments affect Claude and Anthropic, the story is moving quickly and shows no sign of settling soon.
