AI
U.S. Federal Agencies Ordered to Purge DeepSeek AI Over China Data Risks
The Department of Homeland Security has directed civilian agencies to remove the Chinese AI system from government networks and cease new purchases, citing national security concerns about potential access to U.S. user data by Chinese authorities.
TechCrunch — AI
The U.S. Department of Homeland Security has instructed all federal civilian agencies to eliminate DeepSeek AI tools from their IT infrastructure and stop contracting for services built on the Chinese company's technology, citing concerns that sensitive government data could fall under Beijing's purview through China's legal system.
The prohibition extends beyond DeepSeek's direct products to encompass third-party applications and services that incorporate DeepSeek's underlying technology—though agencies may petition for exemptions on a case-by-case basis, according to DHS documentation.
The National Security Rationale#
DHS flagged risks tied to how DeepSeek processes and retains user information, warning that information generated by U.S. government users could be compelled by Chinese law to be disclosed to state authorities. The step reflects broader anxiety over data residency and regulatory overreach in jurisdictions where foreign governments can mandate corporate compliance.
This latest move follows a December 2025 law that already barred DeepSeek deployment in U.S. intelligence and national security agencies. The new order widens that scope to encompass all civilian government departments and their existing systems.
Intensifying U.S.-China AI Competition#
The directive underscores accelerating friction between Washington and Beijing over artificial intelligence dominance. DeepSeek's low-cost, openly released AI models have emerged as serious challengers to established U.S. competitors like OpenAI and Meta since the startup's reasoning model R1 disrupted the AI landscape in January 2025, triggering a notable decline in Nvidia's stock price.
Allegations of Model Theft#
U.S. officials have accused DeepSeek and five other Chinese AI labs—including Alibaba and Moonshot AI—of employing model distillation, a technique that mines outputs from American AI systems to bootstrap their own models. OpenAI and Anthropic have each reported suspected unauthorized extraction attempts, with OpenAI of DeepSeek's suspected distillation activity and detection of intrusions aimed at capturing Claude model outputs.
