Why Combiner calls for legislation on model distillation, saying breaking monopoly begins with open weight freedom

The debate over model distillation techniques is no longer just a technical dispute between research labs; it has become a public confrontation over the future of AI monopoly. While leading closed model labs, notably Anthropic, call for strict regulatory intervention to halt knowledge extraction from their outputs, Gary Tan, CEO of the U.S. accelerator “Why Combiner”, takes the opposite stance, urging regulators not to intervene and to allow emerging labs to practice distillation freely and transparently.

Tan’s stance followed a second report from Anthropic that accused Chinese labs of launching what it described as unlawful distillation attacks, using fake identities and stolen credentials to send intensive queries to their advanced models and extract their reasoning without permission. Dario Amodei, Anthropic’s CEO, had explicitly called for regulatory constraints to curb these practices. However, Tan explained in an interview with CNBC and statements to TechCrunch that he prefers the government not intervene against the technology,instead proposing an American distillation framework that lets local labs extract knowledge from large modelsto build strong open-weight alternatives that outperform Chinese models without needing backdoor infiltration.

Tan’s argument rests on two main points that reveal a double standard among monopolistic model developers. The first point is that restricting customers’ use of information shared by models through APIs constitutes a unjustified limitation, especially since the closed labs themselves trained their systems by sweeping massive amounts of human knowledge and copyrighted content without prior permission. Tan believes that intelligence built on public-interest data should be treated as a public good, not as a product shackled by burdensome service terms.The second point warns that the real nightmare for AI lies in a single commercial company monopolizing the lead, capital, and elite researcherscreating an all-encompassing monopoly that controls the technology’s future.

For tech leaders and founders in the Gulf, Egypt and the Levant, this framing moves the open-weight battle from theoretical debate to a concrete operational choice. Most institutions and startups in the region currently lease inference from closed U.S. APIs, whose terms prohibit using the outputs to train alternative models, keeping the regional ecosystem in a perpetual tenant position and raising monthly compute costs. If distillation is reclaimed as a user right, it would enable Arab tech teams to transfer advanced inference capabilities to smaller, cheaper models hosted locally in their own data centers, protecting data sovereignty and lowering operating costs. It also shifts the skill set demanded in the local market from mere surface-level prompt engineering to designing distillation pipelines and synthetic data for training specialized models.

The equation Tan advocates does not invalidate leading companies’ right to develop profitable business models that cover their massive research costs, but it does set a limit on attempts to stifle competition. Open models remain the sole safety valve to ensure developers worldwide have access to AI tools and prevent this global infrastructure from becoming the exclusive property of a single company.