Full Breakdown
Garry Tan Pushes Back Against U.S. Calls to Restrict AI Model Distillation
By Drooid · · How we work
Core Event – Tan’s Call for Minimal Regulation
At Y Combinator’s annual Demo Day on September 11, 2026, Garry Tan, chief executive of the startup accelerator, told CNBC that regulators should “do nothing” about AI model distillation. He argued that U.S. “open-weight” labs should be free to use the outputs of frontier models through legitimate access, warning that a single dominant proprietary provider would create a “nightmare scenario” for the industry.
Background & Context – U.S. Advisory on Chinese Distillation
The debate follows a joint cybersecurity advisory issued on September 9, 2026 by the NSA, FBI and CISA. The advisory named six China-based AI firms—DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI—as conducting “industrial-scale knowledge distillation” against U.S. frontier models such as Claude, GPT, Gemini and Grok. The agencies warned that such campaigns could shorten AI development timelines and reduce the financial resources needed for competitive model building.
Data & Statistics – Scale of Alleged Distillation
- Anthropic’s threat-intelligence report recorded more than 151 million Claude exchanges between May and July 2026, routed through over 3,500 fraudulent accounts.
- CISA noted that the Chinese firms extracted “billions of tokens” from U.S. models since late 2024.
- Y Combinator’s Demo Day featured 149 machine-learning and AI startups out of 196 presenting companies.
Official Statements & Responses – Government and Industry Positions
CISA Acting Director Nick Andersen urged AI companies to protect their platforms, stating:
- “We strongly urge AI companies to take immediate steps to safeguard their platforms against knowledge distillation campaigns that threaten to close the gap in advancements made by American companies.” — Nick Andersen, CISA
Anthropic’s CEO Dario Amodei called for regulators to crack down on illicit distillation, describing the Chinese campaigns as “illicit” and “fraudulent.” OpenAI has similarly argued that unauthorized distillation amounts to theft of proprietary capabilities.
Conflicting Reports & Gaps – Discrepancies in Exchange Counts
Anthropic’s report cites 151 million Claude exchanges attributed to Alibaba, while its broader summary references nearly 200 million exchanges across five campaigns. Sources do not reconcile these figures, leaving the exact volume of illicit distillation ambiguous.
What’s Next – Potential Policy Directions
The joint advisory signals a possible shift toward stricter enforcement of API-usage policies and increased collaboration among AI providers to detect abnormal prompting patterns. While no regulatory timeline has been announced, agencies suggest future guidance may target both foreign and domestic actors engaged in large-scale distillation. Garry Tan’s advocacy indicates industry stakeholders will continue to lobby for a framework that balances open-weight innovation with protection of frontier-model investments.