WASHINGTON — Top American AI companies are feeding Beijing’s beast and the US military at the same time, a longtime China hawk is warning.
Four major providers of data for AI models — which help collate, label and package the vast troves of info for Anthropic, Google, Meta, OpenAI and the Pentagon — are also quietly working for China’s AI labs on the side, according to a new report.
“America is in an AI race with the Chinese Communist Party,” Rep. Michael McCaul (R-Texas) told The Post.
“We cannot spend billions trying to protect our technological advantage while American companies package up our best human expertise and sell it to companies tied to the Chinese military.”
None of the companies responded to The Posts requests for comment.
China’s top six AI labs have dished out roughly $500 million a year on American data companies, according to a Forbes report.
Four US-based AI data companies reported to have done business with Chinese AI labs are: SurgeAI, Mercor, AfterQuery, and Turing.
McCaul, the former chairman of the House Foreign Affairs Committee, said the companies must stop working for China if they want government contracts.
“If you want to do business with the US government, you should not also be helping a Pentagon-designated Chinese military company build better AI,” he said.
AI requires vast amounts of data so that its models can fine-tune their pattern prediction capabilities. Depending on the type of AI, its training data requirements is often very specialized and hard to obtain.
The data needs to be accurate and labeled properly. Without proper data, tech companies won’t be able to get their AI models to function properly.
SurgeAI, which is generally seen as one of the top data labeling companies in the field, has done work with Chinese firm Tencent, according to Forbes.
SurgeAI has earned some small contracts with the Army and Air Force. But Tencent has been designated by the US as a Chinese military company.
Executives at Chinese firm Tencent, which works on AI, video games, and more, privately described relying on US firm Surge AI, the outlet reported.
Mercor, which reportedly locked down a US government contract last month, is also known to have done business with Tencent. Some 2% of the company’s revenue came from China in the second quarter, a source told Forbes. The firm recruits and organizes experts for AI companies to lean on for training their chatbots.
AfterQuery, which curates data for AI development, reportedly gets $50 million in “recurring revenue” from China, including its dealings with Ant Group — the financial arm of Alibaba.
Lastly, Turing, which helps label and develop specialized training data, has done work with ByteDance, the parent company of TikTok in China.
The US and China are locked in an AI arms race, and Beijing has repeatedly made massive leaps over recent months to narrow the gap.
For years, the US has sought to curtail China’s access to advanced chips, including high-end Nvidia graphics processing units that are widely seen as instrumental to boosting AI.
But China’s access to cleaned-up American AI training data could prove valuable to its efforts to win that race.
The US data companies’ business dealings in China is also raising concerns because compiling and cleaning up data for AI development is known to be a very difficult process.
