Rep. Andrew Garbarino
Rep. Andrew Garbarino speaks at a subcommittee hearing in April 2023. Image: Committee video feed

US companies’ use of Chinese open weight AI draws rebuke from House chairmen

A new congressional report focused on how some American companies are embedding China-developed open weight artificial intelligence models into their tech stacks calls the trend alarming and cautions U.S. firms against the practice.

Pointing to how Huawei and ZTE technology quickly found their place in the marketplace and threatened American networks, the new report warns that if China-developed open-weight models are allowed to take root, removing them down the road will take several years and cost taxpayers billions.

Open weight models’ numerical weights and biases are made public for download. 

The joint investigative report — produced by the House Homeland Security Committee and the Select Committee on the Chinese Communist Party — focuses on Airbnb, DoorDash and Cursor coding platform owner Anysphere as examples of companies deploying Chinese open weight models. The companies were using them for customer assistance, coding projects and taking inventory.

None of the firms immediately responded to a request for comment.

Report co-authors Reps. Andrew Garbarino (R-NY) and John Moolenaar (R-NY) drew the 50-page-long report’s findings from hearings, classified briefings, and agency advisories. The lawmakers also received briefings from DoorDash, Airbnb and Anysphere about how they integrated China-developed open-weight models into their consumer platforms, enterprise software and other systems.

“The investigation uncovered risks that American businesses, policymakers, and national security officials can no longer afford to overlook or downplay,” the report said. “While America is investing billions and racing to build ‘Super Intelligence,’ Beijing is using stolen intelligence to close the gap.”

Silicon Valley startups and Fortune 500 companies alike are adopting Chinese open-weight models because they are cheap, easily customized and work well, according to the report.

But those benefits come with significant negatives, the report said, citing federal investigators’ findings showing that Chinese developers “used proxy networks and fraudulent accounts to harvest outputs from American frontier models and replicate their capabilities.”

The report features 12 policy recommendations, including the imperative to build an American open-weight AI initiative, establish a standard for where models are sourced from, create an AI bill of materials, bar China-developed models from federal contract work and sensitive critical infrastructure functions and evaluate Chinese models systematically.

The report is important because it identifies and explains the risks posed by the U.S. not competing seriously in the low cost open-source model marketplace, according to Mark Montgomery, a cybersecurity expert at the Foundation for Defense of Democracies.

Because China is more limited in the frontier model space — dominated by U.S. firms such as Confluence, OpenAI — its leaders have pursued a strategy to expand global dependence on their open-source AI, using the “traditional Chinese model of state policy and capital” to kill competition, he said.

“Companies and utilities, driven by bottom line resource decisions will, as always, ignore security and safety issues in pursuit of a lower front end cost,” he said. “China knows this recipe well.”

Recorded Future
No previous article
No new articles
Suzanne Smalley

Suzanne Smalley

is a reporter covering digital privacy, surveillance technologies and cybersecurity policy for The Record. She was previously a cybersecurity reporter at CyberScoop. Earlier in her career Suzanne covered the Boston Police Department for the Boston Globe and two presidential campaign cycles for Newsweek. She lives in Washington with her husband and three children.