Garry Tan, chief executive of the startup accelerator Y Combinator, has said he would "do nothing" in response to claims that Chinese artificial intelligence laboratories have distilled the models of OpenAI and Anthropic. Speaking to CNBC at Y Combinator's annual Demo Day, in remarks published on 10 September 2026, Tan argued that regulators should concentrate on preserving a workable balance between open weight and frontier models rather than policing distillation itself.
"We could argue that there should be an American distillation regime," he said. His position places him at odds with frontier developers and some national security figures who have pressed for action against the practice.
What Tan argued
Distillation is the process of using the outputs of a more capable AI model to train a smaller or less capable one, sometimes illicitly. It has been a major point of concern for frontier model developers, namely OpenAI and Anthropic. Tan framed the spread of open weight models as a benefit rather than a threat, provided frontier models keep a price premium that sustains their business model.
"This is actually the ideal case. You want open weight models to give people freedom and access," he said. "If I were a regulator, that's what I would go after." He acknowledged the balance was "a tightrope" but said it "could result in the best possible outcome."
The accusations in the background
Tan's remarks come during an escalating dispute. Anthropic has accused Moonshot AI, DeepSeek and MiniMax of distilling its models. OpenAI believes DeepSeek's V3 and R1 model architectures were distilled from its GPT-4 and GPT-4o models. Neither claim has been independently proven, and both remain allegations.
On 8 September 2026 the National Security Agency, the Cybersecurity and Infrastructure Security Agency and the Federal Bureau of Investigation released an official cybersecurity advisory warning on the topic.
Copyright complicates the complaint
Critics have pushed back on the distillation complaints because much of the data used to train frontier models may itself be covered by copyright law, a point Tan highlighted. The New York Times sued OpenAI and Microsoft in 2023 over unauthorised use of its articles in training data, and a consortium of book authors settled a similar claim with Anthropic in 2025.
Safety: science fact, not science fiction
Tan also advocated a restrained approach to AI safety, arguing that attention should rest on present risks rather than extinction scenarios. "We need to be focused on science fact, not science fiction," he said. His comments follow a period of heightened public anxiety tied to the resignation of Anthropic researcher Jacob Coxon, which coincided with record high fears about existential risk.
He treats cybersecurity as an imminent risk. On 10 September 2026 Anthropic reported that it had blocked Claude access in five cases involving scientists in unspecified foreign countries conducting research on dangerous pathogens, which Anthropic feared could be covert bioweapons work. Tan, by contrast, sees AI-related job losses and economic upheaval as decades away. "It will take decades for this to actually percolate into society, and that's not a bad thing," he said.
Y Combinator's own portfolio reflects its near term focus: of the 196 startups presenting at Demo Day, 149 were categorised as machine learning and AI ventures.
What is established and what is merely claimed
Established: Tan publicly stated he would "do nothing" about distillation and would instead pursue an equilibrium between open weight and frontier models, in remarks published on 10 September 2026. The NSA, CISA and FBI issued a cybersecurity advisory on distillation on 8 September 2026. Anthropic reported blocking Claude access in five pathogen research cases on 10 September 2026. The New York Times litigation against OpenAI and Microsoft was filed in 2023, and the authors' consortium settled with Anthropic in 2025. Of 196 Demo Day startups, 149 were AI ventures.
Merely claimed: Anthropic's accusations that Moonshot AI, DeepSeek and MiniMax engaged in distillation, and OpenAI's belief that DeepSeek's V3 and R1 were distilled from GPT-4 and GPT-4o, remain unproven allegations. Tan's views that job displacement is decades away, and that an open and frontier equilibrium is achievable, are his opinions, not established findings.