Policy
House Report Warns U.S. Companies Against Chinese AI Models
Congressional investigation finds Airbnb, DoorDash, and others embedding China-developed open-weight models despite security concerns.
Key takeaways
- House investigation identified Airbnb, DoorDash, and Anysphere as companies using China-developed open-weight AI models in their operations.
- Lawmakers warn that removing embedded Chinese AI models could take years and cost billions, similar to Huawei/ZTE telecommunications equipment removal.
- Federal investigators found Chinese developers used proxy networks and fraudulent accounts to harvest outputs from American AI models.
- The report recommends twelve policy changes including banning Chinese models from federal contracts and critical infrastructure.
- China is pursuing a strategy to dominate low-cost open-source AI while U.S. companies focus on frontier model development.
Congressional Investigation Targets Chinese AI Adoption
A joint congressional investigation has identified what lawmakers describe as an alarming pattern: American companies are integrating China-developed open-weight artificial intelligence models into their technology infrastructure, potentially creating long-term security vulnerabilities.
The House Homeland Security Committee and the Select Committee on the Chinese Communist Party produced the 50-page report after conducting hearings, classified briefings, and company consultations. Representatives Andrew Garbarino (R-NY) and John Moolenaar (R-NY) led the investigation, which specifically examined how Airbnb, DoorDash, and coding platform Cursor owner Anysphere have deployed Chinese models for customer service, software development, and inventory management.
Open-weight models differ from proprietary systems in that their numerical weights and biases are publicly available for download, making them attractive to companies seeking cost-effective AI solutions.
Why it matters
The report draws a direct parallel to Huawei and ZTE telecommunications equipment, which required years and billions in taxpayer dollars to remove from American networks after security concerns emerged. Lawmakers warn that allowing Chinese AI models to become embedded in critical business systems could create a similar—or worse—dependency that proves extremely difficult and expensive to reverse. The investigation suggests China is pursuing a deliberate strategy to expand global reliance on its open-source AI while American companies remain focused on frontier models.
Security Risks and Stolen Intelligence
According to the report, federal investigators discovered that Chinese developers employed proxy networks and fraudulent accounts to extract outputs from American frontier AI models and replicate their capabilities. This intelligence gathering has enabled Chinese firms to narrow the technological gap while offering products that appeal to cost-conscious businesses.
Mark Montgomery, a cybersecurity expert at the Foundation for Defense of Democracies, noted that China faces limitations in competing with U.S. frontier model developers like OpenAI and Anthropic. Instead, Chinese leaders have deployed "traditional Chinese model of state policy and capital" to dominate the lower-cost open-source segment, betting that companies prioritizing immediate cost savings will overlook security implications.
Policy Recommendations
The report presents twelve policy recommendations aimed at addressing these vulnerabilities. Key proposals include establishing an American open-weight AI initiative, creating sourcing standards for AI models, implementing an AI bill of materials requirement, prohibiting China-developed models from federal contracts and critical infrastructure applications, and developing systematic evaluation processes for Chinese models.
The investigation concludes that while American investment focuses on achieving "Super Intelligence," Beijing is leveraging stolen intellectual property to close competitive gaps through a different market strategy.
The findings were first reported by The Record, which noted that the named companies did not immediately respond to requests for comment on their use of Chinese AI models.