OpenAI is scared of open weight models. Should the US be?

12:33 PM · July 21, 2026

TechCrunch examines why the sudden push in Washington to restrict Chinese made open weight AI models reveals a tension that has more to do with business models than national security. The piece argues that while officials frame the threat from models like Moonshot's Kimi K3 in terms of security risks and alleged intellectual property theft, the more immediate pressure the models create is commercial: a powerful, freely downloadable model undercuts the entire subscription and API based revenue structure that closed labs like OpenAI and Anthropic depend on to justify their enormous valuations and the capital expenditure commitments built on top of them. If organizations can download a model that performs close to the frontier for free and run it on their own infrastructure, the argument goes, the pricing power closed labs currently enjoy erodes regardless of whether the open model was built through any improper means. The analysis is skeptical of how neatly national security and commercial self interest have aligned in the current policy debate, noting that OpenAI's own public statements about the risks of open weight models available to adversaries have grown notably more urgent at almost exactly the moment those models became commercially threatening. The piece concludes that whatever restrictions the US ultimately adopts, they should be evaluated on whether they address a genuine security question rather than simply protecting the market position of the incumbent closed labs that are lobbying hardest for them.