Silicon Valley loves a good panic. Whenever a new competitive threat emerges from overseas, the default reaction from domestic tech giants is usually a frantic call for regulatory walls, emergency export controls, or outright bans. But Y Combinator Chief Executive Garry Tan recently took a radically different stance on one of the industry's most contentious flashpoints, arguing that when it comes to AI model distillation and international copying, the smartest regulatory move is to simply do nothing.
That counterintuitive take cuts straight through the noise of protectionist rhetoric. As frontier labs accuse competitors, particularly in China, of illegally distilling smaller, highly efficient models from massive proprietary systems like OpenAI and Anthropic, the mainstream consensus leans toward tightening the grip. Tan suggests that trying to police model distillation is like trying to stop the tide with a broom. Instead of wasting political capital on unwinnable enforcement battles over intellectual property in a decentralized digital world, the focus needs to shift toward the underlying market dynamics that make frontier models valuable in the first place.
The Reality of Model Distillation
To understand why a prominent tech leader is advocating for a hands-off approach, you have to look at what distillation actually is. In simple terms, distillation is the process where a smaller, cheaper artificial intelligence model is trained using the outputs of a larger, more powerful frontier model. The smaller model learns to mimic the reasoning patterns and capabilities of its parent, yielding high performance at a fraction of the computing cost.
Critics frame this as sophisticated theft. They argue that billions of dollars in infrastructure and research are being undercut when foreign entities harvest proprietary outputs to train lean, low-cost alternatives. National security hawks chime in, warning that giving advanced capabilities to rival jurisdictions threatens western technological dominance.
Yet, treating distillation as a crime misses the technical reality of how software evolves. Information wants to compress. Once a capability exists in latent space, replicating its functional outcome through distillation is an economic inevitability, not a legal anomaly. Trying to stop developers from distilling models is about as effective as trying to ban open-source code repositories.
Why Market Premiums Matter More Than Bans
Tan argues that instead of panicking over copied tech, regulators should focus on maintaining a healthy pricing and performance gap between frontier systems and open-weight models. If you want to stay ahead, legislation isn't the answer—relentless innovation is.
The real magic of the current tech ecosystem isn't locked inside static model weights anyway. It is found in how organizations wire their workflows, manage context, and build execution layers on top of raw intelligence. When you look at how modern startups actually build, the moat is rarely the base model itself. The moat is the product infrastructure, the specific data feedback loops, and the speed of shipping.
If a competitor can distill a model to match last year's flagship capability, your job isn't to sue them or lobby for trade barriers. Your job is to ship a system that makes last year's capability obsolete. Protectionism breeds complacency. Companies that rely on legal moats instead of product velocity eventually stall out.
Refocusing the Safety Conversation
The obsession with model distillation also distracts from actual, pressing risks. Industry debates frequently drift into speculative science fiction scenarios about rogue superintelligence, while everyday vulnerabilities in cybersecurity and infrastructure go under-addressed.
Real harm is happening right now through phishing automation, zero-day exploitation, and corporate espionage. Job displacement is another massive structural shift that will take decades to work through society. Fretting over whether an overseas lab trained a small model on synthetic data generated by an American chatbot is a massive waste of cognitive bandwidth.
Regulators should instead balance the ecosystem by ensuring open-weight models can flourish alongside proprietary frontier systems, keeping the market competitive and dynamic. When barriers to entry drop, the velocity of overall technological progress increases.
Stop treating every technical parity as an existential crisis. Build better products, focus on execution speed, and let the market reward true innovation rather than regulatory capture.
