The US government report on AI has called Chinese models adversary AI. According to the Centre for AI Standards and Innovation (CAISI), Chinese models pose risks to AI developers, consumers, and US national security due to their security shortcomings and censorship. According to the report, Chinese models lag behind their American counterparts in performance, cost, security, and adoption, despite their growing global popularity. The US has had issues with China’s AI, but more so with DeepSeek. China’s most high-profile AI company that challenged the dominance of the US AI. Most recently, DeepSeek has come under fire in the US after being accused of stealing user data and amplifying Chinese state narratives. The US government weighs DeepSeek against its AI models The evaluation by CAISI marks the first time the US government has conducted a comprehensive assessment of DeepSeek’s capabilities and popularity in comparison to top US models, including Anthropic’s Claude Opus 4, OpenAI’s GPT-5, and its open-source model, gpt-oss. The study said that DeepSeek’s models scored lower than US models on almost all of 19 public and internal benchmarks. The report also states that the models are more likely to be jailbroken by hackers and cybercriminals seeking to engage in illegal activities. The Hangzhou start-up’s open-weight models have helped China catch up to the US in the global AI adoption race. Open-weight models, which are also sometimes called open-source models, have their intelligence encoded in variables that are made public. Since January, DeepSeek downloads models on the developer site Hugging Face have gone up by almost 1,000%. Downloads of models from Alibaba Cloud’s Qwen family also went up by 135%, the report found. Alibaba Cloud is closing in on Meta Platforms, the developer behind the Llama family of models, as the second-most popular model of all time. However, US firms still have the most global downloads across all models on the platform, with OpenAI on the lead. More modified models were built on Qwen and shared on Hugging Face than were built by Google, Meta, Microsoft, and OpenAI combined. On the other hand, the study found that OpenAI’s GPT-5-mini costs 35% less on average than DeepSeek’s top model, V3.1, to perform at the same level, using the prices of their application programming interfaces (APIs) as a basis for comparison. However, the report didn’t say that DeepSeek users can install the open-weight models locally. This isn’t possible for users of proprietary US models, who have to pay for API access. Artificial Analysis, a third-party AI testing firm, states that DeepSeek has also released newer models in the past few weeks, lowering official API prices by more than 50% while maintaining the same performance levels. These updated versions were not included in CAISI’s evaluations, which only looked at DeepSeek’s R1, R1-0528, and V3.1 models. DeepSeek is political The report follows the release of the US President Donald Trump’s AI Action Plan in July, which called for evaluating the capabilities and alignment with state narratives of frontier Chinese models. Similarly, the report said that Chinese government filtering was “built directly into DeepSeek models.” Launched in 2023 under the Biden administration, the CAISI was previously known as the US AI Safety Institute before having its name changed when Trump came into office with a new focus on promoting US leadership in AI innovation. On social media, US Commerce Secretary Howard Lutnick stated that his department was helping to ensure “continued US leadership in AI” by publishing these findings. “The report is clear: DeepSeek lags far behind, especially in cyber and software engineering,” Lutnick wrote . “These weaknesses aren’t just technical. They demonstrate why relying on foreign AI is dangerous and shortsighted.” Don’t just read crypto news. Understand it. Subscribe to our newsletter. It's free .
