AI distillation, a practice where developers use the outputs of advanced AI models to train cheaper, competing systems, has exploded into a major policy battleground. What was once a niche technical discussion has become a hot-button issue from Silicon Valley to Washington, D.C., after Chinese lab Moonshot AI released its Kimi K3 model, which quickly proved competitive with offerings from OpenAI and Anthropic. The controversy raises a fundamental question: Is distillation a legitimate efficiency technique or intellectual property theft ?

What Exactly Is AI Distillation, and Why Does It Matter?

At its core, distillation refers to using answers or work products from an advanced AI model to train another model. Think of it like a student reading a textbook, attending lectures, and doing homework, then sharing that completed work with another student who skips the hard work and simply copies the answers. The practice is controversial because it allows developers to create competitive offerings without investing the millions or billions of dollars that companies like Anthropic and OpenAI have spent developing their frontier models.

Google AI lead Jeff Dean first discussed distillation publicly in February 2026, explaining how the technique helps create smaller, more efficient models without relying on one large system. "Through distillation, which is a key technique for making the smaller models more capable, you have to have the frontier model in order to then distill it into your smaller model," Dean said at the time. Five months later, the practice had become a flashpoint in geopolitical AI competition.

How Is Distillation Being Used in the AI Race?

The controversy intensified when White House advisor Michael Kratsios posted on X that Moonshot AI had distilled Anthropic's Fable model to develop Kimi K3. According to Kratsios, Moonshot "developed 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." This allegation frames distillation not as a technical innovation but as a form of intellectual property theft.

Anthropic has already documented similar concerns. In February 2026, the company reported that its Claude model was being distilled "on an industrial scale" by Chinese companies including DeepSeek, Moonshot, and MiniMax, which used approximately 24,000 fake accounts to generate 16 million exchanges. Anthropic characterized stopping illicit distillation as a matter of national security, arguing that its systems prevent state and non-state actors from using AI to develop bioweapons or conduct malicious cyber activities.

Why Are Tech Giants Defending Distillation?

Despite the national security concerns, some of the world's largest technology companies have taken a surprising stance. On Friday, July 25, 2026, Nvidia, Microsoft, Meta, Palantir, and more than 20 other companies released a joint letter urging policymakers to avoid "premature restrictions" on open-weight AI models that would "stifle competition or drive innovation overseas." The signatories argued that distillation is "a widely used technique for model improvement, evolution, and validation".

Box CEO Aaron Levie, one of the signatories, explained the business logic behind the position. "Generally the arc is going to be that the more innovation that there is, whether that's from the U.S. or China or otherwise, you should expect more AI progress, and generally it'll bend toward being even lower cost and more efficient over time," Levie said. The argument reflects a broader concern that overly restrictive policies could slow innovation and push development overseas.

Shashi Bellamkonda, research director at Info-Tech Research Group, noted that distillation is already widely practiced by U.S. companies. Nvidia, for instance, used distillation as part of the training process for its Llama Nemotron series of models. "It is a legitimate and a very valuable technique to train a smaller, cheaper model on outputs of a larger model, and is practiced all the time," Bellamkonda said.

The Hypocrisy Problem: Who Gets to Claim IP Theft?

A significant complication undermines the case that Anthropic and OpenAI are making about IP theft. Both companies have themselves been sued for using copyrighted material without authorization to train their models. Max Pritt, an attorney at Boies Schiller Flexner representing book authors in copyright litigation against AI firms, highlighted this contradiction: "The administration, at least publicly, has focused its efforts on the protection of technology companies' intellectual property, while remaining silent in large part about creators and individuals' intellectual property that was used without authorization".

This inconsistency raises uncomfortable questions about whose intellectual property deserves protection. While the U.S. government has framed Chinese distillation as a national security threat, it has largely ignored the concerns of authors, artists, and other creators whose work was incorporated into AI training datasets without permission or compensation.

How Does AI Affect Trade Secret Protection?

Beyond copyright and patent concerns, the rise of AI distillation raises questions about trade secret protection, a form of intellectual property that protects confidential business and technical information. Trade secrets can cover everything from formulas and algorithms to customer data, pricing strategies, and manufacturing processes. Unlike patents, trade secrets require no registration and can last indefinitely if secrecy is maintained.

However, AI complicates trade secret protection in several ways. When companies submit confidential information to public AI platforms, they risk losing trade secret status. Courts have taken a strict view on voluntary disclosures. In 2025, the U.S. District Court for the Northern District of California held that voluntary disclosure of alleged trade secrets to a consumer-tier AI platform defeated trade secret protection when the platform's terms of service imposed no confidentiality obligation. Similarly, in February 2026, the U.S. District Court for the Southern District of New York reasoned that communications memorialized through a public AI platform were not confidential where the platform was not contractually bound to keep them secret.

Steps to Protect Your Trade Secrets in an AI-Driven World

• Use Enterprise AI Tools: Deploy AI systems locally or under enterprise terms that prohibit training on customer data, restrict human review, impose confidentiality obligations, and provide deletion and audit rights to minimize the risk of unintentional disclosure.
• Implement Strict Governance Practices: Treat AI governance as a trade secret issue rather than merely an information-security or procedural concern, ensuring that your organization maintains reasonable measures to preserve the secrecy of valuable information.
• Protect Prompts and Workflows: Recognize that meticulously crafted prompts, prompt libraries, and internal AI workflows may qualify for trade secret protection if they are not visible to third parties, developed through investment and experimentation, and maintained under internal secrecy controls.
• Evaluate Information Value: Regularly assess whether information you previously regarded as a trade secret remains economically valuable, or whether historical data now has new value because it can be used to train proprietary AI models.

The core legal principle is straightforward: the same information may remain protected in a closed, contractually controlled environment and lose protection if entered into an unrestricted public tool. This means that how you use AI matters as much as what you use it for.

What's at Stake in the Distillation Debate?

The distillation controversy exposes a fundamental tension in AI policy. On one hand, the U.S. government and companies like Anthropic view distillation as a threat to American technological leadership and national security. On the other hand, tech giants and efficiency-focused companies see distillation as a legitimate way to democratize AI and reduce costs. Colin Shea-Blymyer, a research fellow at Georgetown's Center for Security and Emerging Technology, noted that the U.S. government is still trying to figure out its position on the issue.

The stakes are high. If distillation is restricted, it could protect the competitive advantages of U.S. AI leaders but potentially stifle innovation and drive development overseas. If distillation remains unrestricted, it could accelerate global AI progress and reduce costs, but it may also allow competitors to rapidly catch up to American capabilities without bearing the research and development costs. As policymakers grapple with this dilemma, the outcome will likely shape the future of AI competition for years to come.