Beijing-based Moonshot AI released Kimi K3 on July 16 — a 2.8-trillion-parameter open-weights model that topped LMArena's coding leaderboard, the first time a Chinese model has held that spot. It scored 81.2 on the FrontierSWE benchmark, roughly 10 points ahead of GPT-5.6 Sol, and costs 40% less than flagship American models. The announcement landed the same day Xi Jinping was speaking at the World AI Conference in Shanghai; the Nasdaq dropped about 1% as chip stocks sold off. By this week, the Trump administration was openly debating what to do — and the debate broke into a public brawl.

1. Shut It Down (Emil Michael, Pentagon; Anthropic; OpenAI)

Chinese AI is a security risk, and part of its edge comes from stealing U.S. capabilities.

Nobody can control open-weight models once they're out. Dean Ball, OpenAI's head of strategic futures and a former Trump AI adviser, said open-weight models make AI "ungovernable" — his concern is the paradigm itself, not just Chinese models. He warned that open-weight dominance could make AI "full AI communism" — his term for AI becoming state-provided digital infrastructure that no company can profit from.

Part of Kimi K3's edge may have come from stealing. In February 2026, Anthropic accused Moonshot AI, DeepSeek, and MiniMax of creating over 24,000 fraudulent accounts and running 16 million exchanges with Claude to extract its capabilities. Moonshot's share was roughly 3.4 million exchanges, targeting agentic reasoning — exactly what Kimi K3 now excels at.

The White House is moving without waiting for a formal ban. The administration is weighing adding Chinese AI labs to the Commerce Department's Entity List — which would require U.S. companies to obtain a license before using their models — along with federal procurement restrictions and a coordinated public pressure campaign. Top Pentagon official Emil Michael is among the hawks pushing hardest for action.

2. But the Incumbents Want This Ban (David Sacks, PCAST Co-Chair)

OpenAI and Anthropic have been lobbying to ban their open-source competition for years — and now they're close.

Kimi K3 won on price and performance. OpenAI and Anthropic are asking Washington to step in. David Sacks, Trump's former White House AI czar and now co-chair of the President's Council of Advisors on Science and Technology, pushed back Sunday on X. He wrote: "The leading closed labs, already a duopoly in terms of AI model revenue, want the government to eliminate their open source competition." He called what's happening the "weaponization of regulatory uncertainty." Sacks called Anthropic's models "lobotomized" and "woke." Emil Michael, separately, called Ball a "supreme village idiot."

The AI labs have been pitching this ban for years. Leading U.S. AI labs have pitched the administration on banning open-source models roughly every three to five months, according to a source familiar with the discussions. Sriram Krishnan, the White House's senior AI policy adviser, opposed the bans and left the administration in June. After he left, officials pushing for restrictions gained more internal influence.

The weights drop July 27, and enforcement ends with them. Moonshot AI plans to release Kimi K3's open weights on July 27. Developers will be able to run the model locally with no Chinese company involved. Even if the administration adds Moonshot to the Entity List before that date, the model itself will already be out.

Where This Lands

Security hawks in the administration, backed by Anthropic and OpenAI, want to restrict Chinese AI through Entity List designations, procurement rules, and public pressure campaigns. David Sacks and open-source advocates argue those tools would mostly protect OpenAI and Anthropic from competition — and note that a ban is pointless once the weights land on July 27. What the administration decides before that date is the only thing that matters now.

Sources