Why did Representative Ro Khanna ask OpenAI, Anthropic, Google, Meta and SpaceXAI to disclose known attempts by China or other hostile actors to illegally access their AI model weights and explain their cybersecurity safeguards, and how do his concerns about model-weight theft di

Khanna asked the five companies for evidence of attempted unauthorized access and for an account of their defenses because he fears that stolen model weights could give a hostile actor the model itself, undermine the U.S. AI lead and create national-security risks. His request is an inquiry into whether that more direct form of theft has been attempted and how well companies can prevent it—not a claim that China has already stolen their weights.1114

• The distinction: Model weights are a trained model’s internal parameters. Stealing them would mean obtaining the model directly; distillation trains a different model using the original model’s outputs. U.S. officials have accused Chinese firms of distillation, and OpenAI and Anthropic have reported such activity, but Reuters says few cases of actual weight theft are known.413
• Why it matters for safety and competition: Khanna is treating protection of frontier models as both a cybersecurity obligation and a way to limit the spread of powerful AI capabilities. At the same time, he has called for U.S.–China cooperation on AI safeguards; analysts warn that escalating disputes over alleged copying could make such cooperation harder.125
• The policy tension: The Trump administration has proposed voluntary submission of certain advanced models for government cybersecurity testing, while Khanna’s questions press companies to account for their own security practices. The available reporting does not establish that voluntary testing would prevent weight theft, or that any of the five companies’ weights have been stolen.71113