Y Combinator chief executive Garry Tan told CNBC that the United States should create an “American distillation regime” that permits open‑weight AI labs to copy (distill) frontier models from domestic AI companies. The suggestion, reported by TechCrunch on September 11, 2026, is positioned as a response to Anthropic’s recent accusations that Chinese labs are conducting illicit distillation attacks.
What the proposal entails
In the CNBC interview, Tan said, “I would do nothing… We could argue that there should be an American distillation regime.” He elaborated to TechCrunch that the regime would let smaller, American open‑weight labs apply the same training techniques used by leading U.S. frontier‑model developers, thereby creating a broader set of open‑weight options that are not Chinese‑originated.
“He feels it’s an overreach for AI labs to dictate what their customers can do with the information their models share with them… there’s a role government can play here to normalize the fact that access to intelligence that was trained on broad public access data should itself also be more a form of a public good than something locked away behind restrictive terms of service,” Tan told TechCrunch.
The core of the proposal is two‑fold: (1) remove regulatory barriers that prevent U.S. open‑weight labs from distilling frontier models, and (2) treat the resulting intelligence as a public good, akin to publicly funded research.
Why the timing matters
Anthropic, a San Francisco‑based AI firm, recently reported nearly 200 million distillation‑attack queries between May and July 2026, with a majority linked to a campaign traced to Alibaba. The company framed those queries as “illicit distillation attacks” by Chinese labs. Tan’s call arrives while that episode is fresh in the policy conversation, giving his suggestion immediate relevance.
Tan’s position also aligns with a broader industry debate about the balance between open‑weight models—those whose weights are publicly released—and closed‑weight services that restrict downstream use. He argues that restricting API calls to closed‑weight models is “an overreach” and that allowing open‑weight distillation would diversify the U.S. AI ecosystem.
Sector impact and who stands to gain
If policymakers were to adopt such a regime, several groups could be affected:
- Open‑weight startups would gain legal certainty to build on top of frontier models without fearing antitrust or export‑control hurdles.
- Established frontier‑model developers (e.g., OpenAI, Anthropic) might see increased competition from downstream innovators, potentially accelerating feature development.
- Chinese AI labs could lose a covert advantage if their distillation attacks are curtailed by clearer U.S. rules.
- Regulators would need to define the scope of “open‑weight” and decide whether the regime requires licensing, reporting, or other oversight.
What remains unknown is how the U.S. government would operationalize the regime—whether it would be a statutory amendment, an agency guideline, or a voluntary industry standard. The packet does not contain any official response from the Federal Trade Commission, the Department of Commerce, or the White House.
Company background
Understanding the players helps gauge the stakes. The table below summarizes key facts about the three most relevant U.S. AI‑related entities mentioned in the packet.
| Company | Chief executive | Headquarters | Founded | Employees |
|---|---|---|---|---|
| OpenAI | Not provided in packet | Not provided | 2015‑12‑11 | 4,500 |
| Anthropic | Dario Amodei | San Francisco, United States | 2021‑01‑26 | 2,500 |
| Y Combinator | Garry Tan (CEO) | Mountain View, United States | 2005‑03 | Not provided |
| Source: Wikidata entries cited in the research packet; employee counts are as listed in the packet. | ||||
Y Combinator’s role as an early‑stage investor and incubator gives Tan a platform to influence policy debates, while Anthropic’s security findings provide the immediate catalyst for his remarks. OpenAI, though not directly quoted, is a major frontier‑model developer whose open‑weight policies could be reshaped by any new regime.
Potential policy pathways
Several routes could materialize:
- Legislative action: Congress could pass a law defining “open‑weight AI” and setting permissible distillation practices.
- Executive guidance: The White House or a relevant agency could issue non‑binding guidance, similar to the “AI Bill of Rights” draft released earlier this year.
- Industry self‑regulation: A coalition of AI firms could adopt a voluntary code, using Anthropic’s attack data as a baseline for responsible distillation.
Each pathway carries trade‑offs. Legislative routes provide clarity but can be slow; executive guidance is quicker but may lack enforceability; self‑regulation offers flexibility but risks uneven adoption.
What remains unanswered
The packet does not include any statements from U.S. regulators, nor does it contain quantitative estimates of how many U.S. labs would benefit or how much model performance might improve under an open‑weight regime. It also does not specify whether the proposed regime would apply only to domestically trained models or also to foreign‑origin models that are hosted in the United States.
Until those details emerge, the proposal remains a high‑level policy suggestion anchored in a specific security concern—Chinese distillation attacks—as reported by Anthropic.
Outlook
If adopted, an American distillation regime could reshape the competitive dynamics of the U.S. AI sector. Smaller labs would gain a clearer path to innovate on top of frontier models, potentially accelerating the diffusion of advanced capabilities. At the same time, regulators would need to balance openness with safeguards against misuse, a tension that has already surfaced in debates over export controls and AI‑generated content.
For now, the industry watches to see whether policymakers will translate Garry Tan’s interview remarks into concrete rules, and how that translation will affect the broader race between U.S. and Chinese AI actors.