Khanna Probes AI Security, Warns of China's 'Stroke of a Keyboard' Threat

US Representative Ro Khanna is taking a closer look at the security of America's leading AI labs. He has sent letters to the CEOs of OpenAI, Anthropic, Google, Meta, and SpaceXAI, asking them to disclose any attempts by China or other adversaries to steal their model weights. Model weights are the numerical parameters that define a trained AI model, and stealing them could give the thief a significant advantage in AI research.

Khanna's concern is that a successful theft could undermine US leadership in AI. He warned that such a theft could occur 'with the stroke of a keyboard,' implying that it's a relatively simple task. The representative has also flagged distillation as a significant concern, which is when a developer trains a new model using the outputs of an existing system, rather than stealing its weights directly.

This is not the first time Khanna has tackled the issue of China's tactics for acquiring AI capabilities. He held a hearing in April 2026 to examine the country's efforts, and he has also reached out directly to Chinese AI labs, including DeepSeek, Alibaba, and Moonshot AI. Khanna's focus on distillation suggests that lawmakers may eventually look at how US model outputs are used abroad, not just how the weights themselves are protected.