Distillation Under TRUST: The Case for a U.S.-India Trusted AI Model Distillation and Diffusion Corridor

A growing number of U.S. and Indian companies are deploying Chinese open-weight models even as they become flashpoints in U.S.-China technology competition. Washington is weighing punitive measures on Chinese labs accused of large-scale distillation, while Beijing reportedly is considering its own AI export restrictions. However, restrictions will not solve the underlying problem: Developers and companies will still need capable, cheap, and customizable open-weight models. The United States and India should build competitive alternatives under the U.S.-India TRUST initiative—which stands for “Transforming the Relationship Utilizing Strategic Technology”—by establishing a corridor for U.S. and Indian labs to legally distill open-weight models from frontier U.S. models.

Distilling an AI Advantage

When Chinese lab Moonshot AI released its latest Kimi K3 model, independent evaluations found it “comparable” to Anthropic’s Opus 4.8 and OpenAI’s GPT-5.5. U.S. officials quickly accused Moonshot AI of covertly distilling Anthropic’s Fable model. Distillation, or training a smaller model on a larger one’s outputs, is a common practice, but the Trump administration has sought to differentiate between legitimate use and national security concern.

Distillation is unlikely to explain all of China’s progress, but as Sequoia’s Dean Meyer and Konstantine Buhler note, it can “[compress] the costly final gap between a strong base and a near-frontier system,” allowing the student model to reach the frontier using less compute. That creates a powerful competitive strategy for China, which has translated its cost-effective advances at the frontier through distillation to diffuse inexpensive, open-weight models globally. As a result, U.S. technology leadership at the frontier has not translated to leadership among cost-conscious developers, governments, and businesses across the Global South.

India is a case in point. Despite government bans on Chinese apps and restrictions on telecom equipment and other components used in sensitive sectors, such as drones and CCTV cameras, cost sensitivities have pushed Indian consumer technology companies toward capable Chinese open-weight models even as India seeks to develop deeper indigenous capabilities. That same concern is shared in Washington: Neither government wants Chinese technology setting the terms of AI adoption across the developing world.

While these shared concerns have driven the United States and India to collaborate on critical and emerging technology, including AI, across multiple administrations, the United States has proposed only unilateral measures against large-scale distillation and Chinese leadership in open-weight models. Officials have proposed sanctions and entity list designations, or other measures short of a ban that could create “a chilling effect on Chinese open-source tech.”

However, these are defensive responses to a problem that needs a competitive one. Punitive measures do not eliminate the demand that made capable, flexible, and cheap open-weight models attractive in the first place. Nor would U.S. restrictions necessarily prevent Chinese open-weight models from gaining adoption across India and the broader Global South.

Pushing Back with TRUST

A better approach is a collaborative U.S.-India arrangement that plays to each side’s strengths. U.S. labs lead at the frontier but have largely focused on closed models. India wants deeper domestic and open AI capabilities but does not possess the same abundance of frontier-scale compute. Distillation offers a potential bridge. Rather than punishing distillation after the fact, the U.S.-India TRUST should establish a dedicated corridor for U.S. and Indian labs to legally distill safe, flexible, and compute-efficient open-weight models from closed U.S. frontier models.

Such an approach, if carefully structured, can advance shared goals. Indian labs should not be treated as simply receivers of U.S. frontier tech, but as cocreators of low-cost, flexible models suited to the Global South. The United States can credibly claim that the corridor advances the goals of the U.S.-India Roadmap on Accelerating AI Infrastructure, which seeks to accelerate adoption of the U.S. AI stack in India and spur innovations in AI models and AI applications.

The obvious question is why U.S. frontier labs would permit the distillation of their models to create potential open-weight competitors. There are also real intellectual property questions—such as who owns a “student” model trained on a frontier “teacher”—and export control risks that could erode trust, as the U.S. government’s decision to briefly cut off access to Mythos and Fable showed. But a government-enabled corridor can help incentives align. Frontier labs could license legal distillation for a fee, creating a revenue-accreting alternative to the covert theft they are already suffering. Indian labs would still trail the frontier, but could utilize distillation to build and diffuse trusted, capable, compute-efficient models suited to Global South needs.

If the economics prove too hard, the corridor can still set shared standards and rules for safe and legal distillation, exchange best practices on maintaining strong safeguards on open-weight models, and build joint evaluations and audits to create a repository of trusted open-weight models. Either way, these tensions should serve as the negotiating agenda, not as reasons to shelve the idea.

A U.S.-India distillation and diffusion corridor will not be simple to build. But, done right, it would turn a shared vulnerability into shared leverage: India has a pathway to build a near-frontier domestic open-weight ecosystem, the United States sees the U.S. AI stack spread through this trusted ecosystem, and the countries can, together, contest Global South adoption rather than ceding that terrain to Chinese models.

Aman Thakker is a fellow with the Chair on India and Emerging Asia Economics at the Center for Strategic and International Studies in Washington, D.C.