Garry Tan Advocates for US Open-Weight AI Distillation to Counter Monopoly

Garry Tan Proposes "American Distillation Regime" to Prevent AI Monopoly

Y Combinator CEO Garry Tan has called for a shift in how the U.S. handles model distillation, arguing that smaller American open-weight AI labs should be free to distill knowledge from frontier models. Tan suggests that rather than regulating against distillation—a practice often labeled as "illicit" by proprietary labs—the U.S. should normalize it to ensure that cutting-edge intelligence remains a public good rather than a proprietary asset controlled by a single monolithic entity.

This position stands in direct contrast to the stance of frontier labs like Anthropic. In a September 2026 report, Anthropic alleged that Chinese labs are conducting "illicit distillation attacks" using fraud and stolen credentials to extract knowledge from their models, prompting CEO Dario Amodei to call for regulatory crackdowns on the practice.

The Case for Distillation as a Public Good

Tan's argument for allowing distillation is based on two primary premises: the nature of the training data and the risk of market consolidation.

Data Reciprocity and the "Commons"

Tan contends that proprietary AI labs cannot claim a moral high ground regarding the use of their model outputs because the models themselves were trained on vast amounts of public and copyrighted data without explicit permission. He argues that since these models were built by "vacuuming up" human knowledge from the commons, the resulting intelligence should be treated more as a public good than as a product locked behind restrictive terms of service.

Preventing a Monolithic AI Provider

From a strategic perspective, Tan views the concentration of AI power in one or two companies as a "doomer scenario." He warns that if a single company captures the best researchers and the most capital, it could create a monolithic provider that stifles competition and limits access to intelligence. By encouraging a robust ecosystem of open-weight models—even those derived via distillation—the U.S. can maintain a diverse and accessible AI landscape.

Technical and Economic Perspectives on Distillation

Distillation occurs when a model maker uses extensive prompting of a larger "teacher" model to train a smaller, more efficient "student" model. While frontier labs view this as a threat to their business models, other perspectives suggest it is a necessary evolution of the technology.

Efficiency and Resource Conservation

Supporters of distillation argue that it leads to smaller models with comparable capabilities, which increases resource efficiency and reduces the environmental impact of running massive frontier models. Some suggest that distillation is a "transformative work" analogous to Cliff Notes, where the value is extracted and reorganized into a more useful, compact form.

The Economic Moat Challenge

Critics of the distillation-friendly approach question the long-term economics of AI development. If the value of billions of dollars in research and data acquisition can be immediately cloned into open-weight models, there may be less incentive for labs to invest in the expensive process of generating and buying high-quality training data.

Community Insights and Counterpoints

Discussion among technical communities highlights a deep divide between those who view distillation as a right and those who see it as a breach of contract.

Arguments in Favor of Distillation

• Fair Use: Many argue that distillation is a form of fair use or reverse engineering. One contributor noted, "In every other area of manufacturing and tech I can use a machine to build a new machine that competes with the original machine."
• Anti-Competitive Behavior: Some view the restriction of distillation as an attempt by incumbents to protect a "non-existent moat" through FUD (Fear, Uncertainty, and Doubt).
• Net Neutrality for Intelligence: There is a suggestion that AI access should be guaranteed, and banning distillation is akin to forcibly leaving certain developers behind.

Arguments Against Distillation

• Terms of Service (ToS) Integrity: Some argue that "illicit distillation" involves breaking contracts, using IP proxies, and fraudulent accounts, which undermines business trust.
• Bias and Transparency: Concerns have been raised that distilling "black box" models inherits the undisclosed ideological biases of the original providers, potentially tainting the open-weight ecosystem with the biases of a few powerful interests.

"The nightmare scenario, the doomer scenario for AI is that there’s just one company... It has the best access to capital. It has the best AI researchers. It runs away with it and suddenly there’s one company that’s monolithic. And that would be bad." — Garry Tan