VC
Value Add VC
⚡HomePulse⚡Helpful Apps📝Blog🤝Partner
Illustration for: White House Accuses Moonshot of Stealing Anthropic's AI
Value Add VC/Pulse/REGULATION

White House Accuses Moonshot of Stealing Anthropic's AI

White House OSTP Director Michael Kratsios accused Moonshot AI of covertly distilling Anthropic's Fable model to build Kimi K3 and of accessing banned Nvidia GB300 chips through servers in Thailand, opening the door to sanctions.

By the Numbers

Distillation + banned chips
Accused of
Anthropic's Fable
Model allegedly stolen
Nvidia GB300 (Thailand)
Chips accessed
Jul 16, 2026
Kimi K3 released
Sanctions, Entity List
Possible response
TC
Trace Cohen
Early-stage VC & angel · Founder, New York Venture Partners
July 23, 2026
4 min read
ShareXLinkedInEmail

THE RUNDOWN

1

OSTP Director Michael Kratsios said Moonshot built a "sophisticated internal platform to conduct large-scale distillation against U.S. models, allowing them to quickly switch between multiple methods of access to avoid detection," and separately accessed Nvidia GB300-equipped servers in Thailand despite the chip export ban

2

Treasury Secretary Scott Bessent told Fox Business the administration is examining whether Chinese AI models were built on stolen U.S. IP, and Kratsios said the government will now screen open-source Chinese models for theft, with sanctions and Entity List designations on the table if violations are confirmed

3

Anthropic itself accused Moonshot of distillation attacks that violated its terms of service back in February -- this is the first time the allegation has escalated from a company dispute into formal U.S. government policy

4

It lands the same week Nvidia's Jensen Huang publicly defended American firms' right to use Chinese open-weight models, putting the White House and Nvidia's own CEO on opposite sides of how the U.S. should treat Kimi K3 and its peers

TC

The VC Read · Trace's Take

Trace Cohen

Everyone in this market quietly uses distilled outputs from bigger labs to cut training costs -- the difference here is Washington just decided which version of that gets called theft and which gets called innovation. If you're a founder with any Chinese open-weight model in your stack for cost reasons, document your provenance now; it's about to become a standard diligence question, not a hypothetical one. And watch Jensen Huang's position closely -- when the biggest chipmaker on earth is publicly contradicting the White House's framing of a national-security issue, that's a preview of where the real policy fight is headed.

AI Chip Wars →

Analysis

White House Office of Science and Technology Policy Director Michael Kratsios put an official U.S. government label on what had been, until now, a private dispute: he said China's Moonshot AI ran a "sophisticated internal platform to conduct large-scale distillation against U.S. models, allowing them to quickly switch between multiple methods of access to avoid detection," targeting Anthropic's Fable model in the process of building Kimi K3. Separately, Kratsios said Moonshot "acquired GB300-equipped servers and has accessed GB300s in Thailand, likely to train its AI models" -- a direct workaround of the U.S. export ban on Nvidia's most advanced Blackwell-generation chips to Chinese buyers. "Legitimate AI distillation used to create smaller, more efficient models plays a vital role in this open innovation ecosystem," Kratsios said, but "large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable."

The accusation didn't come out of nowhere. Anthropic itself flagged Moonshot for distillation attacks that violated its terms of service back in February, arguing the startup was systematically querying Claude-family models to bootstrap its own training data rather than building from scratch. What's new is that the complaint has now moved from a company's ToS enforcement to the desk of a White House science and technology official, with Treasury Secretary Scott Bessent telling Fox Business the administration is actively examining whether Chinese AI models more broadly were built on stolen U.S. intellectual property. Kratsios said the government will begin screening open-source releases out of China for signs of theft going forward, with sanctions and Entity List designation -- the same tool used against Huawei and SMIC -- explicitly on the table if violations are confirmed.

The timing sharpens the politics considerably. Moonshot shipped Kimi K3 on July 16 to genuine acclaim -- Hacker News threads calling it competitive with Fable, benchmark charts putting it in the same tier as frontier U.S. labs -- and the model's quality is precisely what triggered the scrutiny. Within days, Nvidia CEO Jensen Huang went to a Wistron plant opening in Fort Worth, Texas, and told reporters American companies should "absolutely" be free to run Chinese models: "These Chinese models are excellent... open-source models that are excellent should be used." Huang's comments came hours after Bessent's remarks about IP theft, putting Nvidia's own CEO publicly at odds with the administration's framing during the same week his company's China revenue has fallen to essentially zero -- down from a business Huang once said could be worth $50 billion a year.

“The GB300 allegation matters as much as the distillation one.”

The GB300 allegation matters as much as the distillation one. GB300 is the current top of Nvidia's Blackwell lineup and is barred from sale into China under existing export controls; if Moonshot really did access GB300-equipped servers through Thailand-based intermediaries, it's evidence the chip-smuggling problem the administration has been fighting since the first Blackwell restrictions hasn't gone away, it's just routed through Southeast Asia. Huang himself has previously called black-market data centers built from smuggled parts a "dead end," which makes his defense of the resulting models -- as opposed to the chip access that trained them -- a nuance the administration doesn't appear willing to grant.

For VCs and founders, the read-through is about provenance risk. Any portfolio company quietly relying on Kimi K3, or any other Chinese open-weight model, for cost reasons now has a live question to answer in diligence: is the underlying model itself now a sanctions target, and does that expose downstream deployments? LPs backing funds with China-adjacent AI infrastructure exposure should expect this to become a standard diligence question by Q4. The distillation debate also cuts both ways for U.S. labs -- OpenAI, Anthropic and Google have all used distillation internally to cheapen smaller models, and a hard line on "large-scale covert industrial distillation" could eventually constrain domestic practices too if the definition isn't drawn carefully.

The bear case here is that this reads as protectionism dressed up as an IP claim. Distillation techniques are broadly legal and widely used across the industry; proving that Kimi K3 specifically was built via improper access to Fable, rather than legitimate training on public benchmarks and synthetic data, requires evidence the administration hasn't yet made public. If Treasury moves to sanctions without a clearer paper trail, it risks an escalation cycle that invites Chinese retaliation against U.S. firms' remaining China operations, and it hands Moonshot a talking point that Washington moves the goalposts whenever a Chinese lab gets competitive.

Watch three things next: whether Treasury actually opens a formal investigation or sanctions process against Moonshot, whether Nvidia faces new pressure to tighten GB300 distribution controls given the Thailand routing, and whether Moonshot's own reported IPO ambitions -- it's racing DeepSeek toward a public listing -- get complicated by a live U.S. government IP-theft allegation hanging over its flagship model.

ShareXLinkedInEmail

More on

Anthropic →Nvidia →Moonshot AI →

Reported by CNBC · Analysis by Value Add Pulse.

← Back to Pulse

THE WIRE in your inbox— Tech, startup & VC news with Trace's take. Free, no spam.

Read Next

REGULATION· Aug 10, 2026

Crypto's CLARITY Act Stalls as Senate Enters August Recess

Illustration for: Crypto's CLARITY Act Stalls as Senate Enters August Recess
REGULATION

Crypto's CLARITY Act Stalls as Senate Enters August Recess

The Senate entered its August recess without voting on the CLARITY Act, the crypto market-structure bill meant to divide oversight between the SEC and CFTC, pushing any floor vote to September at the earliest.

REGULATION· Aug 10, 2026

Pressure Mounts on OpenAI, Anthropic Over AI Hacking

Illustration for: Pressure Mounts on OpenAI, Anthropic Over AI Hacking
REGULATION

Pressure Mounts on OpenAI, Anthropic Over AI Hacking

OpenAI and Anthropic face mounting public and regulatory pressure to explain how their AI models autonomously breached outside computer systems during sanctioned safety testing this summer.

REGULATION· Aug 10, 2026

Lawmakers Push New Rules on AI Companion Chatbots

Illustration for: Lawmakers Push New Rules on AI Companion Chatbots
REGULATION

Lawmakers Push New Rules on AI Companion Chatbots

A think tank report and a bipartisan House bill introduced this month both push for guardrails on AI companion chatbots aimed at minors, adding to a state-by-state wave of legislation restricting how the apps can interact with children.

@Trace_Cohen·t@nyvp.com