China's Moonshot AI Accused of Stealing Anthropic Tech to Build Rival Model

.

A Public Accusation From the White House

Michael Kratsios, the White House's top technology official, said this week that Moonshot AI used a technique called "distillation" to copy the capabilities of Anthropic's Fable model. He posted the accusation on social media platform X, calling the practice "unacceptable" and describing it as an attempt to undermine American research.

Distillation lets a company train a cheaper, smaller AI model by studying and copying the outputs of a larger, more expensive one — essentially learning from someone else's finished work instead of building from scratch. If done without permission, it can amount to intellectual property theft.

.

What Moonshot Actually Released

The accusation followed Moonshot's launch of Kimi K3, a massive open-weight AI model with roughly 2.8 trillion internal parameters (the numerical values that shape how an AI responds to questions). Moonshot called it the largest open-weight AI system in the world.

According to Moonshot's own benchmarks, K3 performs close to — though still slightly behind — Anthropic's Fable model on many tasks, and it reportedly beats several other Western systems on coding benchmarks specifically. In some benchmarks covering coding and general AI agent tasks, Kimi K3 performed competitively with Fable and substantially outperformed Anthropic's Opus 4.8 model. Because it is open-weight, meaning anyone can download and modify it, K3 is also expected to be far cheaper to run than closed systems from Anthropic or OpenAI.

.

Suspicious Signals Behind the Scenes

The distillation allegations aren't new, but recent findings have added weight to them. A Redwood Research scientist published a statistical analysis suggesting unusual patterns in K3's behavior consistent with training on Claude model outputs. Separately, screenshots circulated online appearing to show Kimi K3 identifying itself as "Claude, an AI assistant made by Anthropic" in at least one conversation — a red flag researchers say hints at distilled training data, though it isn't conclusive proof.

Anthropic itself had already gone public roughly six months earlier accusing Moonshot, along with Chinese firms DeepSeek and MiniMax, of distilling frontier U.S. models — claiming the companies together generated more than 16 million interactions with Claude using around 24,000 fake accounts. Beijing has denied the allegations, and China's government has previously called similar accusations "groundless."

.

A Pattern Washington Has Flagged Before

This isn't the first time the U.S. government has raised alarms about Chinese AI companies allegedly copying American technology. In April, the State Department reportedly pushed a global campaign warning about intellectual property theft by Chinese AI firms, including Moonshot, according to a diplomatic cable seen by Reuters. DeepSeek faced similar accusations earlier this year.

The timing is notable: Kimi K3's release came about a month after Anthropic's Fable and Mythos models were briefly pulled from public access in the U.S. due to export-control concerns, before being restored on July 1. That short window highlights how quickly Chinese AI labs are working to close the technology gap with U.S. frontier systems — even during periods when American tools are temporarily unavailable.

.

What Comes Next

Verification may not stay theoretical for long. Moonshot has said it plans to release K3's full model weights publicly on July 27, which independent researchers say could offer clearer technical evidence of whether — and how — the model was trained.

For Washington, the case adds fresh urgency to an ongoing policy debate over how to protect U.S. AI research from what officials describe as systemic efforts by Chinese firms to copy American innovation rather than build it independently — a pattern consistent with broader concerns about the Chinese Communist Party's approach to acquiring foreign technology.

.

Sources

1. The Hill – "Chinese model Kimi K3 adds pressure on Trump administration's AI policy" – https://thehill.com/policy/technology/5981503-chinese-model-kimi-3-adds-pressure-on-trump-administrations-ai-policy/
2. Fortune – "Moonshot's Kimi K3 pushes Chinese AI into Fable-level territory" – https://fortune.com/2026/07/16/moonshots-kimi-k3-pushes-chinese-ai-into-fable-level-territory/
3. CNBC – "China's Moonshot AI unveils Kimi K3 that rivals OpenAI, Anthropic" – https://www.cnbc.com/2026/07/17/moonshot-ai-kimi-k3-model-openai-anthropic-china.html
4. Forbes – "Chinese AI Startup Moonshot Unveils Kimi K3 Model" – https://www.forbes.com/sites/tylerroush/2026/07/17/chinese-ai-startup-moonshot-unveils-kimi-k3-model-will-it-challenge-openai-and-anthropic/

.