The launch of Kimi K3, a free open-weight model from Chinese startup Moonshot that rivals the capabilities of OpenAI and Anthropic’s best systems, has fractured the Trump administration’s AI policy circle into openly warring factions with no agreed strategy for how to respond.
The model, released July 17, stunned Silicon Valley and Washington by demonstrating frontier-level performance at zero cost to the user. Within days, Moonshot temporarily halted new subscriptions after Kimi overwhelmed its compute capacity. The wake-up call hit an administration whose AI advisors have spent months arguing in opposite directions.
Two irreconcilable visions
David Sacks, the former AI and crypto czar who left office earlier this year, publicly dismissed concerns about Chinese AI dominance. He described Anthropic’s models as “lobotomized” and “woke,” arguing that Chinese AI is popular precisely because it has fewer restrictions. “US AI companies want the government to eliminate their open-source competition,” he wrote on X, advocating deregulation as the correct response.
Emil Michael, a top Pentagon official, took the opposite position, calling OpenAI’s new head of strategic futures a “supreme village idiot” for suggesting a softer approach. He insisted that “the democratic process, not some Deep State scheme” should determine the US response to Kimi, signaling support for stronger government controls.
Dean Ball, a former Trump AI advisor now at OpenAI, criticized the new White House AI review process as a “de facto licensing regime for frontier AI.” He predicted that rather than imposing direct bans, the administration would use soft power to make US companies afraid to adopt Chinese models like Kimi.
“The Trump AI world is at war with itself,” observed Anton Leicht of the Carnegie Endowment. He noted that Chinese AI models have already rattled US stocks, posing “a threat for an administration that really doesn’t want more economic bad news.”
Policy paradox
The internal conflict stems from a structural contradiction in the administration’s approach. Trump loosened export controls on AI chips to China, allowing Nvidia to sell more hardware in exchange for a government cut. Yet the administration also announced a crackdown on model distillation in April, after OpenAI and Anthropic complained that Chinese companies were training on their outputs.
The result is that Chinese labs have access to Nvidia chips and US model outputs, two ingredients US policy was designed to restrict, and have used them to build a free competitor that threatens the business models of the very companies Washington was trying to protect.
Kimi’s arrival also reignites a debate the administration has never resolved. Sacks sees it as evidence that deregulation works: fewer restrictions produced a better model. Michael and others see it as evidence that stronger government control is needed. There is no sign of a unified policy response emerging.
Meanwhile, the public has little sympathy for OpenAI and Anthropic, which spent much of 2026 defending their copyright practices and pricing models. A Pew Research study in June found growing distrust of AI companies generally. An AI model that is free, open, and nearly as capable as the best US systems presents an uncomfortable political calculus.
Sources: China’s AI models have Trump’s AI world at war with itself (MIT Technology Review, July 2026)