A new federal order is pushing DeepSeek out of civilian government systems across the United States. The move targets not only DeepSeek’s own products, but also tools built by third parties using the Chinese startup’s technology, turning data security concerns into a direct government-wide restriction.
The Department of Homeland Security announced the ban on Monday, directing federal civilian agencies to stop using DeepSeek AI tools. The order focuses on the risk that sensitive information could move to China through the models, whose servers are located there, along with cybersecurity weaknesses that could expose government systems to attack.
A Government-Wide Ban With Clear Boundaries
The order applies to federal civilian agencies, including the Treasury, State, and Justice departments. It excludes the Department of Defense, because the Department of Homeland Security can direct civilian agencies but does not control the military or intelligence services.
The timing adds another layer to the policy. DHS originally issued the order on July 24, 2026, and the Federal Register published it on October 19, 2026. Executive agencies can request exceptions within 60 days, creating a short window for departments that argue they need to keep access to DeepSeek tools.
The restriction also fits into a broader series of actions against the company’s models across the U.S. government:
- Congress restricted DeepSeek use in intelligence community systems in December 2025.
- U.S. Commerce Department bureaus told staff that DeepSeek was banned on government devices in March 2025.
- The Defense Information Systems Agency moved to block access to DeepSeek models in January 2025.
Together, these measures place limits on DeepSeek across civilian agencies, intelligence systems, and defense networks, though the new DHS order itself covers federal civilian systems.
Data Transfers and Vulnerabilities Drive the Decision
The central concern is where sensitive government data might go when employees use DeepSeek’s AI models. The order states that using the tools may result in sensitive U.S. government data being transferred to China, raising concerns about potential access by Chinese authorities.
Security officials also point to vulnerabilities inside the technology. DeepSeek’s AI models present multiple threat vectors, and the models may contain malware or other security weaknesses. Open-source research identified multiple critical vulnerabilities in DeepSeek’s iOS app, adding a software security problem to the wider debate over data control.
That combination gives the federal action a wide reach. The order does not stop at DeepSeek-branded products; it also covers tools created by third parties using DeepSeek. Any agency assessing an AI system must therefore consider both the tool’s name and the technology operating underneath it.
DeepSeek has been contacted for comment. DeepSeek and China’s Embassy in the U.K. did not respond to requests for comment at the time of the report.
DeepSeek’s Rise Meets Washington’s Security Line
The government’s decision follows the rapid attention drawn by DeepSeek’s R1 model, which the startup released last year. R1 drew comparisons to models from OpenAI and Google, while its development with reduced-capacity chips and at a lower cost made the system a major point of interest in the AI industry.
That rise now faces a powerful counterforce: government rules built around where data travels, who may access it, and whether the software contains exploitable weaknesses. The same model features that attract users can create new questions for agencies handling sensitive information.
Chinese foreign ministry spokesperson Mao Ning said, “China consistently opposes the over-generalization of national security as well as the politicization of science, technology and trade issues.” Her statement places the U.S. restriction inside a wider dispute over technology, security, and trade.
The 60-day exception period will show whether any civilian agencies can justify continued use of DeepSeek tools. At the same time, the ban signals that federal AI adoption will depend on more than capability or cost. Data location, system vulnerabilities, and links to foreign technology now sit at the center of the government’s decision.
DeepSeek’s R1 helped prove that lower-cost AI models could command global attention. The federal order shows the next contest will be just as important: whether those models can earn enough trust to operate inside the systems that protect sensitive government data.
Based on
