US executives warn of security risks from Chinese AI models as Washington weighs sanctions

US executives warn of security risks from Chinese AI models as Washington weighs sanctions Anthropic's CEO calls the trajectory 'dystopian' while DeepSeek shows a 94% jailbreak success rate in testing Share Add us on Google by Editorial Team Jul. 25, 2026 The AI cold war just got a lot warmer. Top executives from Anthropic and OpenAI are sounding alarms about Chinese AI models that are rapidly catching up to, and in some cases matching, their paid US counterparts, all while carrying what they describe as serious security vulnerabilities and potential ties to Beijing’s censorship apparatus.

Anthropic CEO Dario Amodei used the word “dystopian” to describe where things could head if Chinese AI models continue spreading without regulation. Advertisement The security gap is not subtle Models from firms like DeepSeek and Moonshot AI’s Kimi are now competitive with paid US offerings in areas like coding and cybersecurity tasks. During security testing, DeepSeek models showed a 94% success rate for jailbreaks.

Compare that to US models, which have invested heavily in making their systems harder to exploit, and the gap becomes glaring. US officials are also raising concerns about intellectual property theft, alleging that some Chinese firms may have built their models using improperly acquired US technology. Treasury Secretary Scott Bessent announced on July 21 that the US would assess Chinese open-source models for potential IP theft, with sanctions on the table as a possible consequence.

Congress is already digging in The US House Homeland Security Committee and the China Select Committee launched a joint probe back in April, investigating the use of Chinese AI models by American companies including Cursor and Airbnb. Both committees continued pressing for answers into July about why US firms are integrating foreign AI models that could create national security vulnerabilities. Some jurisdictions aren’t waiting for a federal answer.

Multiple US states and government agencies have already banned the use of DeepSeek, treating it as a known risk rather than a hypothetical one. Broader federal restrictions are actively being considered, though the exact form they would take remains in flux. Disclosure: This article was edited by Editorial Team.

For more information on how we create and review content, see our Editorial Policy. AI US executives warn of security risks from Chinese AI models as Washington weighs sanctions Anthropic's CEO calls the trajectory 'dystopian' while DeepSeek shows a 94% jailbreak success rate in testing by Editorial Team Jul. 25, 2026 Share Add us on Google The AI cold war just got a lot warmer.

Top executives from Anthropic and OpenAI are sounding alarms about Chinese AI models that are rapidly catching up to, and in some cases matching, their paid US counterparts, all while carrying what they describe as serious security vulnerabilities and potential ties to Beijing’s censorship apparatus. Anthropic CEO Dario Amodei used the word “dystopian” to describe where things could head if Chinese AI models continue spreading without regulation. Advertisement The security gap is not subtle Models from firms like DeepSeek and Moonshot AI’s Kimi are now competitive with paid US offerings in areas like coding and cybersecurity tasks.

During security testing, DeepSeek models showed a 94% success rate for jailbreaks. Compare that to US models, which have invested heavily in making their systems harder to exploit, and the gap becomes glaring. US officials are also raising concerns about intellectual property theft, alleging that some Chinese firms may have built their models using improperly acquired US technology.

Treasury Secretary Scott Bessent announced on July 21 that the US would assess Chinese open-source models for potential IP theft, with sanctions on the table as a possible consequence. Congress is already digging in The US House Homeland Security Committee and the China Select Committee launched a joint probe back in April, investigating the use of Chinese AI models by American companies including Cursor and Airbnb. Both committees continued pressing for answers into July about why US firms are integrating foreign AI models that could create national security vulnerabilities.

Some jurisdictions aren’t waiting for a federal answer. Multiple US states and government agencies have already banned the use of DeepSeek, treating it as a known risk rather than a hypothetical one. Broader federal restrictions are actively being considered, though the exact form they would take remains in flux.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.