Senators cited concerns that DeepSeek’s R1 model could leak U.S. user data or aid Chinese military operations.
The letter requests an investigation into potential backdoors, data flows to China, and licensing violations.
Some experts warn open-source models may expose enterprises to security risks even without direct state involvement.
A group of Republican lawmakers has asked the U.S. Commerce Department to investigate potential national security threats tied to DeepSeek, a Chinese company behind an open-source AI model known as R1.
In a letter sent August 1 and made public on Tuesday, the senators cited risks that the model could leak U.S. user data and aid China’s military and surveillance efforts.
Led by Senator Ted Budd (R-NC), the letter was co-signed by Senators Bill Cassidy (R-LA), John Cornyn (R-TX), Marsha Blackburn (R-TN), Todd Young (R-IN), John Husted (R-OH), and John Curtis (R-UT).
DeepSeek “has willingly provided and will likely continue to provide support to China’s military and intelligence operations,” the letter reads, citing official assessments and describing the claims as “deeply troubling” due to potential risks of data exposure and misuse.
The lawmakers warned that DeepSeek R1’s open-source design allows developers to access and modify its model weights, creating potential vectors for abuse. They pointed to the model’s ability to generate harmful content and its release without comprehensive safety testing as further cause for concern.
“Model weights can also be publicly downloadable, which users will be facing scrutiny for remote code execution vulnerabilities, theft, tampering, or exploitation,” Chris Anderson, CEO of ByteNova AI, told Decrypt. “When provenance and auditability are unclear, enterprises risk unknowingly exposing sensitive data or enabling adversarial misuse.”
Ensuring that such applications are “secure and not prone to leaking secure information and malign exploitation is paramount,” the senators wrote, calling AI development for business and consumer use a key front in the competition with China.
The Trump administration had previously weighed banning DeepSeek in April over concerns that China would gain a competitive edge.
One example cited in the letter involved R1 producing instructions for a social media campaign encouraging self-harm among teenagers, as well as guidance for constructing a bioweapon.
The senators argued that such outputs highlight the dangers of deploying permissively licensed AI models without proper oversight.
Still, ByteNova’s Anderson acknowledged that restrictions in government systems may be warranted, but warned of broader tradeoffs.
In terms of global AI development, bans “will lead to an oligopoly among American AI companies, thereby slowing down some AI innovations and causing issues,” he said, noting how those issues also connect with the challenge of decentralizing AI innovation outside of the U.S.
“Federal environments demand a precautionary principle: even a low-probability breach could have catastrophic consequences for national security,” Anderson added.
A weekly AI journey narrated by Gen, a generative AI model.
