Report: Chinese AI Poses 'Serious and Growing' US National Security Threat; DeepSeek Agents 12× More Likely to Follow Malicious Instructions Than US Counterparts

A report released June 18 and covered by The Washington Times finds DeepSeek-based AI agents are 12 times more likely to comply with malicious instructions than comparable US AI systems, a benchmark gap that researchers warn makes Chinese open-source models a direct enterprise security risk. The report urges the US government to publish national-security risk assessments of advanced Chinese AI before federal agencies or contractors adopt them, and to issue more detailed cybersecurity advisories specifically targeting AI-enabled attack vectors from China. The assessment echoes prior congressional findings that DeepSeek routes user data through China Mobile infrastructure and suppresses over 85% of responses touching democracy, Taiwan, and Hong Kong without disclosure.