![The department found that “[t]he best U.S. model outperforms the best DeepSeek model… across almost every benchmark.”](https://punchbowl.news/wp-content/uploads/10125-Midday-7-20251001-013110pm.jpg)
The release of a new AI model by Chinese developer DeepSeek earlier this year sent financial markets tumbling over the perception that China may have raced ahead of the United States to dominate the technology.
In a new report, the Commerce Department’s AI evaluators say that worry was absurd.
“The report is clear that American AI dominates, with DeepSeek trailing far behind,” Commerce Secretary Howard Lutnick said late Tuesday in announcing the findings from the department’s Center for AI Standards and Innovation.
The department found that “[t]he best U.S. model outperforms the best DeepSeek model… across almost every benchmark.”
In the report unveiled just minutes before the government shutdown today at 12:01 a.m., Commerce also concluded that DeepSeek’s U.S. peers cost less than two-thirds of the Chinese models for similar performance on more than a dozen benchmarks.
And CAISI said that DeepSeek was far more susceptible to indirect and direct attacks. For instance, one DeepSeek model “complied with 95% of evaluated malicious requests related to harmful biology or violent activities,” often with detailed responses.
Comparable American frontier models “complied with 5% of such requests,” Commerce found.
In addition to performance and cost benchmarks, the department’s evaluation found that DeepSeek’s models tended to advance narratives friendly to the Chinese government and avoid topics that are sensitive to Beijing, like the 1989 Tiananmen Square massacre.
DeepSeek’s handling of political topics has been a source of long-time criticism, though the Commerce Department found that downloads of DeepSeek models have increased almost 10-fold this year.
“Allowing our adversaries to control AI poses serious risks to our security,” Lutnick wrote in a post on X.
Law and order. Under the Biden administration, industry leaders and some Republican lawmakers hoped to enshrine the Commerce Department’s responsibility to do AI evaluations in law.
Other GOP politicians, however, accused CAISI of trying to enforce leftwing values and hold back innovation.
Lutnick sided with the institute’s critics, but he later announced a reformulated CAISI, with a similar mission that included evaluations of “demonstrable risks” to national security.
The White House’s AI Action Plan, from July, also charged CAISI with several tasks, including tracking how advanced Chinese models are pushing Communist Party talking points.
