Rep. Ro Khanna, the ranking Democrat on the House Select Committee on the Chinese Communist Party, is demanding that five leading American artificial-intelligence companies disclose known attempts by China or other hostile actors to gain unauthorized access to their closely guarded AI model weights, intensifying Washington’s scrutiny of whether America’s most valuable AI technology is adequately protected. Letters sent to OpenAI, Anthropic, Alphabet, Meta and SpaceXAI seek information about attempted and successful intrusions, cybersecurity defenses and measures protecting model weights—the numerical parameters containing much of a trained model’s acquired capabilities. Khanna warned that successful theft could allow China to bypass years of expensive American research and potentially undermine the United States’ technological advantage. The inquiry follows similar concerns raised earlier this year by Republican Sens. Jim Banks and Chuck Grassley, underscoring bipartisan anxiety over Chinese espionage targeting frontier AI. Although American companies have reported Chinese developers using techniques such as model distillation to learn from the outputs of U.S. systems, publicly documented cases involving the outright theft of frontier model weights remain limited.
Key Takeaways
- Khanna has asked OpenAI, Anthropic, Alphabet, Meta and SpaceXAI to disclose known Chinese or other hostile attempts to access proprietary model weights and explain how effectively those assets are protected.
- Model weights present an unusual national-security vulnerability because stealing them could provide an adversary with much of the capability of an already-trained AI system without requiring the enormous computing expense and research necessary to independently reproduce it.
- Congressional concern is bipartisan: Republican senators previously questioned nine major AI companies about Chinese espionage, personnel vetting, insider threats, privileged access to model weights and policies for notifying the federal government about suspected foreign compromises.
In-Depth
America’s artificial-intelligence competition with China increasingly involves more than developing better models. It also requires preventing Beijing from acquiring the technology American companies have spent billions of dollars creating.
Rep. Ro Khanna is pressing five leading U.S. AI companies to disclose attempted or successful efforts by China and other hostile actors to obtain proprietary model weights. These numerical parameters embody much of what an AI system learned during training. Stealing them could potentially allow an adversary to acquire sophisticated capabilities without duplicating the enormous investment required to train the original model.
The concern extends beyond conventional corporate intellectual-property theft. Advanced models increasingly possess capabilities relevant to cybersecurity, military planning, surveillance and scientific research. Consequently, inadequate protection of frontier AI technology could become a national-security vulnerability rather than merely a commercial loss.
The issue has also crossed party lines. Republican Sens. Jim Banks and Chuck Grassley previously demanded information from nine AI companies about Chinese espionage, insider threats, personnel vetting and access to sensitive AI assets. Their inquiry specifically asked whether companies could robustly protect model weights from Chinese actors and whether they would notify Washington following suspected compromises.
There is an important distinction between stealing model weights and model distillation. American developers have accused Chinese companies of using outputs from advanced U.S. models to improve competing systems, but publicly established instances of outright model-weight theft remain scarce.
Washington therefore faces a legitimate security challenge without an easy solution. Protecting frontier technology requires strong cybersecurity and counterespionage measures while preserving the private innovation responsible for America’s AI leadership in the first place.
Sources
- https://www.theepochtimes.com/us/us-ai-giants-asked-to-disclose-any-chinese-attempts-to-obtain-closely-guarded-model-weights-6097557
- https://www.reuters.com/legal/litigation/leading-democrat-asks-ai-firms-data-any-chinese-access-sensitive-code-2026-10-01/
- https://www.banks.senate.gov/news/press-releases/senator-banks-raises-concerns-over-chinese-espionage-targeting-u-s-artificial-intelligence-sector/
- https://docs.house.gov/meetings/ZS/ZS00/20260416/119165/HHRG-119-ZS00-Wstate-MahmoodY-20260416.pdf
- https://www.intelligence.senate.gov/2026/05/20/intelligence-authorization-act-for-fiscal-year-2027-as-reported-on-may-20-2026/