U.S. Senator Marsha Blackburn (R-Tenn.), along with Senator Ted Budd (R-N.C.) and five other Senate colleagues, has called on U.S. Secretary of Commerce Howard Lutnick to investigate DeepSeek Artificial Intelligence for its alleged ties to China’s military and intelligence operations. The senators sent a letter urging the Department of Commerce to examine these allegations and assess potential risks from Chinese open-source AI models that may be transmitting information to servers linked to the Chinese government.
The letter states: “We write to you regarding concerning security vulnerabilities and the potential compromising of American personal and enterprise data through the use of DeepSeek Artificial Intelligence (AI) reasoning models. Recent reporting states that U.S. officials believe that DeepSeek ‘has willingly provided and will likely continue to provide support to China’s military and intelligence operations.’ The article further states that U.S. officials allege that DeepSeek is sharing user information and statistics with Beijing’s surveillance apparatus. These allegations are deeply troubling. DeepSeek’s R1’s model release in late January demonstrated the aptitude of People’s Republic of China (PRC) national AI talent and the progress their home-grown models have made relative to leading U.S. products. The Trump Administration has rightly emphasized winning the AI competition against the PRC, and the development of AI use case applications for businesses and consumers is an important facet of that competition. Ensuring that such applications are secure and not prone to leaking secure information and malign exploitation is paramount.”
Concerns highlighted by the senators include reports suggesting that DeepSeek’s R1 model may not have undergone sufficient safety testing before its release, which could allow it to generate harmful content or be exploited for malicious purposes. For example, a Wall Street Journal reporter reportedly prompted R1 into generating dangerous instructions, while Wiz Research identified a publicly accessible database belonging to DeepSeek containing sensitive internal data.
In their letter, the senators wrote: “It is probable that R1 did not undergo comprehensive red-teaming and safety tests to prevent the generation of harmful content prior to release. For example, a Wall Street Journal reporter was able to get R1 to write text for a social media campaign intended to encourage self-harm amongst teenage girls, as well as to provide instructions for carrying out a bioweapon attack. Shortly after R1’s release, Wiz Research found a publicly accessible database belonging to DeepSeek, which allowed full control over database operations including the ability to access internal data. Concerningly, Wiz researches found over a million lines of log streams containing sensitive information like chat history and secret keys. Given their restricted access to the most advanced compute resources, the PRC has seemingly adopted a strategy of embedding open-source AI models into applications and services as a way to compete with the U.S. for global AI leadership… The U.S. government has previously recognized the threats posed by the wide-spread adoption of certain PRC technologies. For example, Congress funded a multi-billion program to remove Huawei telecommunications hardware from American networks after it was determined that such hardware could contain backdoors for PRC espionage. In order to prevent a similar situation, we ask that you identify and evaluate any potential backdoors or vulnerabilities posed by Chinese open-source models like DeepSeek’s R1… We also request that you provide a briefing to members of Congress and relevant Congressional Committees on your findings and any threats posed by Chinese open-source models.”
The senators’ letter asks Secretary Lutnick not only for an investigation but also requests details on how cybersecurity efforts will be prioritized within initiatives like those led by the Center for AI Standards and Innovation (CAISI). They further urge examination into whether such AI models might bypass export controls related to semiconductors.
The correspondence was co-signed by Senators Bill Cassidy, M.D., John Cornyn, John Husted, John Curtis, and Todd Young.
