US authorities are investigating whether Chinese AI startup DeepSeek obtained and utilised restricted US-made AI chips in developing its models. The probe comes despite export controls intended to limit China’s access to advanced semiconductor technology, according to a report in Reuters. The investigation is part of broader US efforts to regulate the flow of high-performance AI hardware to China, which Washington views as a strategic competitor in AI and semiconductor development.
The Department of Commerce is assessing whether DeepSeek acquired high-performance Nvidia chips, such as the H20, through indirect procurement channels. US regulations prohibit the sale of these processors to China due to national security concerns, but restricted hardware may have been sourced through intermediary nations, including Singapore and the UAE. Investigators are also reviewing potential supply chain loopholes that could enable unauthorised access to restricted technology, particularly through resellers or offshore entities operating in jurisdictions with less stringent enforcement of US export rules.
DeepSeek, founded in 2023, has rapidly expanded in the AI sector with the launch of its R1 assistant earlier this month. The application has gained traction and recently outperformed leading US-developed AI models in download rankings. The Chinese company has confirmed using Nvidia’s H800 chips, which were legally available in 2023, but its potential access to restricted components remains under scrutiny. The US has previously taken action against Chinese firms suspected of circumventing export restrictions, raising questions about whether DeepSeek may have obtained advanced AI chips through similar channels.
US lawmakers have urged stricter enforcement of export controls to prevent illegal access to advanced AI technology. House Select Committee leaders John Moolenaar and Raja Krishnamoorthi have called on National Security Advisor Mike Waltz to assess the security risks of restricting AI chip exports to DeepSeek.
“DeepSeek made extensive use of Nvidia’s H800 chip, the first chip that Nvidia designed specifically to fall outside US export controls,” the lawmakers wrote. “This demonstrates what the Select Committee has long argued: frequently updating export controls is imperative to ensure the PRC will not exploit regulatory gaps and loopholes to advance their AI ambitions.”
DeepSeek is also facing intellectual property concerns following allegations from OpenAI. The US-based AI developer claims DeepSeek may have used proprietary technology to develop its own models. OpenAI asserts that DeepSeek employed a technique called “distillation,” where a smaller AI model is trained using outputs from a larger one.
The rise of DeepSeek has triggered volatility in the technology sector, with investor concerns about the ease with which the firm built a complex AI assistant at minimal expense influencing market movements. Following the release of its AI assistant, major US technology stocks declined, including a significant drop in Nvidia’s share price. Market reactions also reflect broader concerns among investors about increased competition between US and Chinese AI firms.
Cybersecurity firm uncovers DeepSeek database exposure
The company’s data handling practices have also come under scrutiny, particularly its collection of user inputs and chat histories. DeepSeek stores this data on servers in China, raising security concerns. The US Navy recently issued an advisory against using the application, citing potential risks.
Earlier this week, cybersecurity firm Wiz Research reported that a misconfigured database exposed DeepSeek’s sensitive data online. The publicly accessible database contained user chat histories, API keys, and internal system details without authentication. After being alerted, DeepSeek secured it. Wiz Research identified about 30 internet-facing subdomains linked to DeepSeek, mostly for chatbot interfaces and API documentation. However, two open ports, 8123 and 9000, provided direct access to the unsecured database. The firm stated that the misconfiguration left sensitive records vulnerable to unauthorised access.