Garry Tan AI Distillation: US Labs Strategy & Impact

Y Combinator CEO Garry Tan has ignited a significant debate within the artificial intelligence community by advocating for US open-weight AI labs to actively ‘distill’ knowledge from American frontier models. This controversial stance, articulated in recent interviews, directly challenges calls from some leading AI labs, like Anthropic, for regulators to crack down on such practices. Tan’s proposal aims to foster a more robust domestic open-weight AI ecosystem and counter the growing reliance on Chinese open models.

By SarmayaNext AI & Emerging Tech Desk • ✓ Fact-Checked • Published September 2026

What is Garry Tan’s Proposal for US Open-Weight AI Distillation?

Y Combinator CEO Garry Tan has publicly urged US regulators to avoid intervening in the practice of AI model distillation, instead suggesting the establishment of an ‘American distillation regime.’ This regime would encourage smaller, American open-weight AI labs to utilize distillation techniques on models developed by US frontier AI labs, such as OpenAI, Anthropic, Meta, and xAI. Tan’s objective is to cultivate a diverse and robust set of American open-weight AI options, reducing the current dependence on Chinese alternatives like DeepSeek and Qwen.

Distillation is a legitimate and common AI training technique where a ‘student’ model learns from the outputs and reasoning of a larger, more capable ‘teacher’ model. This process typically involves extensively prompting the frontier model and using the resulting question-answer pairs to train a smaller, more cost-effective model that can mimic much of the larger model’s performance. While effective, it has become a contentious practice. Anthropic, for instance, recently released a second report alleging that Chinese labs are engaging in ‘illicit distillation attacks,’ involving hidden identities, fraud, and stolen credentials to bypass permissions. Anthropic CEO Dario Amodei has previously called for US regulators to intervene and restrict distillation, a position directly opposed by Tan, who clarifies he is not advocating for illicit methods but for open, permissioned access.

Strategic Implications of AI Distillation for US Competitiveness and Open-Weight Models

Garry Tan’s advocacy for an American distillation regime stems from a strategic imperative to bolster US AI competitiveness. He argues that Chinese labs, including DeepSeek and Alibaba’s Qwen team, successfully closed a multi-year gap with Silicon Valley by employing distillation tactics. By encouraging American open builders to distill from models like Llama, GPT, Claude, and Grok, Tan believes the US can rapidly develop its own high-performing open-weight models, thereby breaking the current reliance of US startups on Chinese open-source options.

Tan’s argument is twofold: he views it as an overreach for frontier AI labs to dictate how customers use information obtained through API calls to closed-weight models. He points out that proprietary AI labs themselves did not seek permission when ingesting vast amounts of human knowledge, including copyrighted material, to train their foundational models. Tan asserts 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,’ suggesting a role for government in normalizing this perspective.

Beyond competitiveness, Tan frames the proliferation of open-weight models as a safeguard against a potential ‘doomer scenario’ where a single, powerful proprietary provider controls the immense power of frontier AI. He states, ‘The nightmare scenario, the doomer scenario for AI is that there’s just one company.’ This perspective aligns with Y Combinator’s significant interest in the AI sector, as it funds hundreds of AI startups building on frontier models and was an early supporter of OpenAI. Tan also redirects the broader AI safety debate, urging policymakers to focus on immediate, tangible risks like cybersecurity and infrastructure security, rather than ‘Hollywood-style doomsday scenarios,’ citing the OpenAI-Hugging Face hack as a pertinent example of current vulnerabilities.

The debate highlights a fundamental tension between intellectual property rights, national strategic interests, and the desire for an open, accessible AI ecosystem. While some, like David Sacks, have previously labeled distillation as IP theft, Tan’s reframing positions it as a necessary tactic for American innovation and a counter to the perceived dominance of Chinese open models. This discussion will likely shape future regulatory approaches and the competitive landscape of global AI development.

The Insider Take

Garry Tan’s proposal underscores a critical strategic pivot for the US AI sector: embracing a ‘fight fire with fire’ approach to counter China’s rapid advancements in open-weight models. This move, if adopted, could redefine intellectual property norms in AI, shifting the balance from strict proprietary control towards a more open, competitive ecosystem driven by national interest. The core challenge lies in distinguishing legitimate distillation from illicit practices, a distinction that will require careful policy calibration to avoid stifling innovation while protecting against fraud.

Frequently Asked Questions About Garry Tan AI distillation

What is AI model distillation, and why does Garry Tan support it for US labs?

AI model distillation is a technique where a smaller ‘student’ model learns from the outputs of a larger, more capable ‘teacher’ model. Garry Tan supports it for US labs to rapidly develop homegrown open-weight AI options, reduce dependence on Chinese models, and foster a diverse AI ecosystem, viewing it as a legitimate and strategic competitive tool.

How does Tan’s view on AI distillation differ from other AI leaders like Anthropic?

Tan advocates for an ‘American distillation regime’ and minimal regulatory intervention, arguing against overreach by proprietary labs. In contrast, Anthropic’s CEO Dario Amodei has called for regulators to crack down on distillation, particularly citing ‘illicit distillation attacks’ by Chinese labs involving fraud and stolen credentials.

What are the broader implications of this debate for AI intellectual property and national competitiveness?

The debate highlights a tension between protecting proprietary AI models and fostering an open ecosystem for national competitiveness. Tan argues that intelligence trained on public data should be a public good, and that allowing distillation is crucial for the US to keep pace with global AI advancements, particularly against Chinese models.

“I would do nothing. We could argue that there should be an American distillation regime.” — Garry Tan

“Controlling what users and customers do with API calls to closed weight models feels constraining, and 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.” — Garry Tan

“They are at the frontier and driving it forward. We want that to be fundable, and be a great business model ongoing. You want open weight models to give people freedom and access.” — Garry Tan

“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.” — Garry Tan

“I think that we need to be focused on science fact, not science fiction. I saw ‘Terminator 2’ also. It’s a great movie.” — Garry Tan

“If there was a breach and a coordinated attempt by agents to take over our infrastructure, what do we do about it? Do we know where a given agent is running? What data center? How do we shut it off?” — Garry Tan

PS: For educational and informational purposes only. Technology specifications and availability are subject to regional rollout and device compatibility.