Security News

US Congress Warns Firms Against Building Chinese Open Weight AI Into Their Systems

The Record · 21 Nov 2026
Key Takeaway Before adopting any open weight AI model, find out where it comes from and record it, so you can replace it quickly if its origin becomes a risk.

A new report from two US House committees, Homeland Security and the Select Committee on the Chinese Communist Party, has called the growing use of China-developed open weight AI models by American companies alarming. It cautions US firms against building them into their technology. Open weight models are those whose numerical weights and biases are published for anyone to download.

The 50-page report names Airbnb, DoorDash and Anysphere, the owner of the Cursor coding platform, as examples. According to the report, the companies used the models for customer assistance, coding projects and taking inventory. The authors, Reps. Andrew Garbarino and John Moolenaar, drew on hearings, classified briefings and agency advisories, and received briefings from the three companies. None of the firms immediately responded to a request for comment.

The report compares the trend to the spread of Huawei and ZTE technology, which it says quickly found a place in the market and then threatened American networks. It warns that if Chinese open weight models take root, removing them later will take several years and cost taxpayers billions. It says companies are adopting them because they are cheap, easy to customise and work well. It also cites federal investigators' findings that Chinese developers used proxy networks and fraudulent accounts to harvest outputs from American frontier models and replicate their capabilities.

The report makes 12 policy recommendations. They include building an American open weight AI initiative, setting a standard for where models are sourced from, creating an AI bill of materials, barring Chinese models from federal contract work and sensitive critical infrastructure functions, and evaluating Chinese models systematically. Mark Montgomery of the Foundation for Defense of Democracies said companies driven by bottom line decisions will tend to ignore security in pursuit of a lower upfront cost.

Why it matters

Businesses often treat a free, downloadable model as a simple component, but the report frames model origin as a supply chain risk that is costly to reverse once embedded in customer service, coding and inventory workflows. The assumption that low cost and good performance are enough to justify adoption is being challenged, and the call for an AI bill of materials suggests customers and regulators may soon ask what models sit inside your products.

What to do

  • Inventory every AI model your business and its software suppliers use, including where each model was developed.
  • Ask vendors whether their products embed China-developed open weight models.
  • Record model sources in an internal AI bill of materials.
  • Check contract and government supply requirements that may restrict Chinese-developed models.
Answering to more than one framework? Run them as one security and AI governance programme ->

Summarised by CISO AI from The Record, written with Claude Sonnet 5.5. We link back to every original so you can read it yourself.