White House Accuses Moonshot of Stealing Anthropic AI

What happened

A senior White House official has accused Chinese AI startup Moonshot AI of improperly copying technical work from Anthropic's most advanced model, known internally as Fable, to build its newly released Kimi K3 system 13410. The allegation was delivered publicly by Michael Kratsios, a top Trump administration technology official, who said Wednesday that Moonshot both distilled Fable's outputs to train K3 and obtained restricted Nvidia chips it should not have had access to under U.S. export controls 410. Reuters reporting carried by KELOLAND and Seeking Alpha adds detail on the chip side of the accusation, saying Moonshot gained access to advanced Nvidia hardware, with one account specifying Nvidia GB300 chips allegedly routed through Thailand to skirt sanctions 310.

The claim lands at a sensitive moment in U.S.-China AI competition, as Washington has spent more than two years tightening export controls on advanced semiconductors while Chinese firms have raced to close the gap through cheaper, more efficient model-training techniques. "Distillation" — training a new model by having it learn from the outputs of a more powerful rival system — has become one of the central flashpoints in that rivalry, since it lets a challenger approximate the performance of a frontier model without matching its computing investment.

Context: Anthropic's parallel rise

The Moonshot accusation surfaces alongside a run of other Anthropic news that underscores how central the company has become to the AI industry's infrastructure and talent battles. AMD announced it will invest as much as $5 billion in Anthropic as part of a broader computing-power arrangement, a deal that also includes a chip supply agreement aimed at helping AMD challenge Nvidia's dominance in AI hardware 89. CNBC frames the investment as one more entry in a string of infrastructure agreements Anthropic has signed this year to expand its compute capacity 9.

Anthropic has also been drawing attention for reasons unrelated to chips or geopolitics. The company hired Tom Blomfield, cofounder of British bank Monzo, who is taking a leave from Y Combinator to join Anthropic's compute team, adding to a roster of high-profile recruits 7. Separately, a new Claude advertisement titled "There's hope in hard questions" has proven polarizing, with some viewers calling its imagery haunting and dystopian while others found it moving; the ad reportedly drew commentary even from OpenAI's Sam Altman 56. And in the investment world, Motley Fool has weighed in on how Anthropic stacks up against OpenAI as both companies approach trillion-dollar valuations, suggesting only one of the two is currently delivering results that match its hype 2.

Where the reporting agrees

Every outlet covering the Moonshot allegation agrees on the core facts: a Trump administration official publicly accused the Chinese startup of distilling Anthropic's Fable model to produce Kimi K3, and separately accused Moonshot of accessing Nvidia chips restricted from export to China 13410. Outlets also agree the accusation was framed as part of a broader U.S. concern about Chinese circumvention of chip export controls, and that it was delivered by a senior White House technology official rather than by Anthropic itself 410. On the business side, CNBC and MarketWatch independently confirm the same AMD-Anthropic figure — an investment of up to $5 billion tied to a computing and chip arrangement — with no discrepancy between the two accounts 89.

Where it doesn't

The accounts diverge mainly in specificity rather than substance. Only Seeking Alpha's account names the exact Nvidia chip in question, the GB300, and describes a routing path through Thailand used to evade sanctions — a level of detail not repeated in the KELOLAND/Reuters or Hill accounts, which describe the chip access more generally as "restricted" or "banned" Nvidia hardware 3410. It is also worth noting that the entire episode rests on an accusation from a U.S. official rather than on independent technical verification of distillation; none of the reporting indicates that Anthropic or an independent research body has confirmed the claim, and Moonshot's own response is not detailed across the coverage. That gap matters because distillation allegations in AI are notoriously hard to prove from outside a company's training pipeline, and the coverage so far reflects a government characterization rather than a settled technical finding.

The bottom line

Taken together, the coverage shows Anthropic sitting at the center of two very different currents in the AI industry at once. On one side, it is a target: the alleged victim of intellectual-property appropriation by a fast-moving Chinese rival, in a dispute now amplified by the White House itself. On the other, it is a magnet for capital and talent, landing billion-dollar infrastructure deals and marquee hires even as its advertising choices spark public debate. The Moonshot allegation itself remains, on the evidence reported so far, an official accusation rather than a proven fact — serious enough to shape policy discussion, but not yet independently substantiated in the reporting.