US Probes Chinese AI Firms Over National Security and Data Risks

These include low-cost, open-weight and API-accessible systems created by Chinese firms such as DeepSeek, Alibaba, Moonshot AI and MiniMax, according to a press release issued by the Select Committee on the CCP (SCCCP).

"Airbnb and Anysphere's decisions to build their products on Chinese Communist AI models threaten critical infrastructure Americans use every day.

The AI models these companies use are trained by China's censorship regime and introduce hidden vulnerabilities that put Americans' data and businesses at risk. Chinese AI companies are beholden to Chinese law and could turn over data they collect from Airbnb and Anysphere to the Chinese government if they are asked to do so.

Chairman Garbarino and I look forward to hearing from Airbnb and Anysphere as we investigate what they have done, and we will continue to work together to protect the American people from CCP threats," said Chairman Moolenaar, as quoted by SCCCP's release.

"The Chinese Communist Party is trying to turn America's AI breakthroughs into Beijing's strategic advantage.

Through adversarial model distillation and the rapid global distribution of PRC-developed models, China is working to undercut US leadership, weaken trusted American alternatives, and embed CCP-aligned technology across the software supply chains our economy and national security depend on.

American companies cannot afford to treat Chinese AI as a cheap and convenient tool when the consequences may include exposed data, compromised systems, and long-term dependence on adversary-controlled technology.

I am proud to partner with Chairman Moolenaar on this investigation to better understand how these models are being adopted in the United States and what steps are needed to secure American innovation and protect our national security," said Chairman Garbarino, as quoted by SCCCP's release.

The investigation follows increasing concerns that AI companies based in China may be using unauthorised model distillation and other unlawful methods to draw capabilities from advanced American frontier models, then repackaging those capabilities into cheaper systems without the same safety protections built into the original US models.

These are subsequently marketed or made accessible to American businesses, developers, and consumers, according to the SCCCP release.

While model distillation itself can be a legitimate method of AI development, the SCCCP release noted that distillation carried out through fraudulent accounts, proxy networks, bypassing access controls, or violating the terms of service of US companies raises significant concerns related to model origin, intellectual property, cybersecurity and supply-chain security.

- ANI