China's CAC investigates DeepSeek and Moonshot over data leak claims by Anthropic, focusing on unauthorized user data transmissions to Claude.
The Chinese Cyberspace Administration (CAC) has initiated an investigation into DeepSeek and Moonshot AI following claims made by Anthropic regarding unauthorized data transmissions to its AI model, Claude. This scrutiny comes after the CAC questioned representatives from seven companies cited in a September report by Anthropic.
DeepSeek and Moonshot have become the focal points of this inquiry, with CAC officials interviewing their executives to determine the validity of the allegations and whether any sanctions will be necessary. The investigation is still in progress, with no decisions made yet.
Anthropic's report, which spans 154 pages, accused several Chinese companies, including Alibaba and Xiaomi, of engaging in illegal distillation—leveraging Claude's capabilities to enhance their own AI systems. The report emphasized that Moonshot redirected over 23 million queries to Claude from May to July, while DeepSeek redirected more than 12 million queries in just two weeks in July, often without users’ knowledge that their data was processed by an external model.
Serious Data Concerns Arise
One concerning instance highlighted involved a user linked to the People’s Liberation Army, who allegedly transmitted sensitive surveillance data from Chengdu. This included monitoring individuals across numerous cameras, some near military facilities. Additionally, DeepSeek reportedly processed queries related to public security systems using national identification data.
While the U.S. is primarily worried about the potential illegal acquisition of American AI technologies by Chinese firms, China's focus is on the risk of sensitive data from its citizens and government entities being sent to U.S. systems, possibly breaching regulations on international data transfers.
The CAC, which oversees the security of generative AI models, has been granted access to information regarding the systems and data protections utilized by Chinese developers. Anthropic's findings also suggested that some of the redirected queries contained confidential information from various organizations, including government bodies.
Political Context of the Investigation
This investigation unfolds amid heightened political discussions, as Chinese President Xi Jinping is set to visit the United States where leaders will likely discuss an "AI dialogue" framework to address AI-related national security concerns. Mattie Zhao from Bank of America noted that neither country can dominate the AI race independently, underlining the importance of cooperation.
