White House Split on Chinese AI Strategy as Moonshot's Kimi K3 Rivals US Leaders

The Distillation Standoff

The Trump administration faces a deepening internal rift over how to counter China’s artificial intelligence advances, particularly following Moonshot AI’s release of Kimi K3, a model that reportedly matches the capabilities of leading systems from Anthropic and OpenAI. According to Wired AI, the split pits White House officials seeking aggressive restrictions against the Commerce Department, which views such measures as impractical to implement.

The immediate trigger is a distillation attack—a technique in which Chinese AI labs train their own models using outputs from American systems to accelerate development. According to Wired AI, the White House announced that Moonshot built Kimi K3 by distilling Anthropic’s Claude Fable 5 model. Treasury Secretary Scott Bessent characterized such covert distillation as “IP theft” in remarks to Fox Business, signaling that sanctions could follow.

White House Pushes for Presidential Action

Senior White House officials have begun discussing formal measures to prevent Chinese AI labs from conducting distillation attacks against U.S. models, according to Wired AI’s sources. This policy push gained urgency after Anthropic last month alleged that Alibaba executed the “largest known distillation attack” against it to date. According to Wired AI, presidential action is expected, though it is not currently anticipated to take the form of an executive order.

The White House has elevated distillation attacks to a top-tier national security concern. The administration’s earlier focus on preventing Claude Mythos and Fable 5 from revealing vulnerabilities in government and critical infrastructure has been eclipsed by the emergence of open-weights Chinese models—systems with minimal safeguards that can be downloaded and deployed by hostile actors globally.

Commerce Department’s Alternative Vision

The Commerce Department, which oversees export controls through its Bureau of Industry and Security, has emerged as a counterweight to White House hard-liners. According to Wired AI, no formal request for input on new restrictions has yet reached the Commerce Department, which harbors its own policy preferences.

Commerce Secretary Bessent has explored middle-ground approaches, including creating financial incentives for American AI labs to develop and release their own open-weights models as a counterbalance to Chinese alternatives. Bessent has held recent conversations with leaders at multiple AI labs to discuss this strategy. According to Wired AI, Bessent is straddling a regulatory middle ground—he imposed export controls on Anthropic—yet remains more permissive than some colleagues.

Why This Matters

The outcome of this intra-administration dispute will shape whether the U.S. pivots toward enforcement-heavy restrictions on AI knowledge transfer, or toward competitive market dynamics. If the White House prevails and implements sanctions or legal barriers against distillation, Chinese labs will face real constraints but may shift tactics toward alternative training methodologies. If Commerce Department reasoning dominates, the U.S. strategy shifts from blocking to accelerating domestic open-source development—a fundamentally different approach to competitive advantage. Teams building AI infrastructure and policy compliance frameworks should monitor whether formal presidential action emerges and what form it takes; the distinction between executive order and regulatory guidance carries significant operational implications for labs working across jurisdictions.