Defending Against DeepSeek: Congress’ Federal Firewall

A NATO ally has branded China’s AI chatbot as a Trojan horse for Beijing’s spies. The Czech Republic’s cybersecurity agency issued a warning on July 10 that led to an immediate ban of DeepSeek from its government devices. The move comes amid growing international alarm over China’s use of AI tools to harvest sensitive information for state intelligence operations.

Adversarial AI as a National Security Threat

The Czech Republic’s National Cyber and Information Security Agency’s (NÚKIB) technical analysis found that DeepSeek “almost certainly” collects and stores user data — including chat history, device details, and other highly sensitive information — raising serious questions about its data-sharing practices. NÚKIB also raised concerns about China’s legal framework, warning that its various national security laws coerce Chinese firms to hand over data. NÚKIB also noted that DeepSeek’s founder was “probably involved in research on dual-use technologies” — reinforcing concerns about the company’s potential links to China’s military applications and suitability for government use.

The Czech Republic’s ban reflects a growing international consensus. U.S. allies and partners, including Australia, South Korea, Italy, Taiwan, and the Netherlands, have issued bans or similar warnings on government use of DeepSeek. Several U.S. agencies, including the U.S. Navy and NASA, have also taken steps to prohibit its use.

Congress Wants To Ban Foreign Adversary-Linked AI From Government Use

U.S. lawmakers are also focused on the national security risks posed by DeepSeek. A recent report by the U.S. Select Committee on the Chinese Communist Party identified DeepSeek as a “profound threat” to U.S. national security and called for immediate congressional action to improve export controls on, and address risks from, Chinese AI models.

In late June, Congress introduced the bipartisan No Adversarial AI Act to limit federal use of AI tools developed by China, Russia, Iran, and North Korea — except in narrowly defined cases such as national security, research, training, and other mission-critical functions. Designated offices would be tasked with creating and updating a public list of AI tools “produced or developed” by listed adversaries. These tools would then be banned from federal use. Supporters of the bill argue these measures would establish a “clear firewall” between adversarial AI systems and U.S. government data.

However, Congress’ efforts to restrict federal use of DeepSeek may be undercut by recent policy shifts. This week, the Trump administration rolled back export restrictions on Nvidia. This shift would allow sales of powerful H20 chips to China — potentially accelerating DeepSeek’s development and strengthening adversarial capabilities.

Risks and Roadblocks to Implementation

While these restrictions could strengthen federal cybersecurity and data practices and reduce reliance on adversarial-developed AI, effective implementation will require addressing significant challenges.

The Trump administration’s aggressive cuts to the federal government’s technical talent have created a “severe shortage” of staff with AI expertise, limiting the government’s capacity to research and identify high-risk systems or manage exception waivers effectively.

In addition, the exemption process itself risks becoming an operational bottleneck. Agencies seeking waivers would need to submit written petitions — a step that could delay mission-critical work, given the limited experts to vet the submissions. A similar problem occurred with the federal ban on Chinese-manufactured drones, where officials from multiple agencies reported delays to research and operations due to waiver-related wait times.

What Agencies Must Do Next

The federal government should ensure adequate AI expert staffing and technical capacity to evaluate and mitigate risks posed by adversarial AI tools. A lack of sufficient in-house expertise could hinder legitimate government research and effective use of powerful AI tools. Strengthening the federal workforce will be critical to identifying high-risk AI systems and implementing appropriate restrictions.

Congress should also consider ways to improve interagency efforts to maintain continuity for research and mission-critical work. Exemption processes should be designed with clear timelines and streamlined pathways to ensure smooth, efficient ways to prohibit the use of adversarial AI tools in the federal government.

Leah Siskind is the Director of Impact and an AI Research Fellow at FDD. Meera Hatangadi is an intern at the Center on Cyber and Technology Innovation (CCTI) at FDD. For more analysis from FDD and CCTI, please subscribe HERE. Follow FDD on X @FDD and @FDD_CCTI. FDD is a Washington, DC-based, nonpartisan research institute focusing on national security and foreign policy.