What happened
RAND's Center for the Geopolitics of Artificial General Intelligence examines whether technical counter-distillation measures — raising the cost of distillation or degrading the quality of distilled models — can meaningfully protect the U.S. frontier AI advantage against rivals training new models on the outputs of more capable ones. The authors, Michael J.D. Vermeer and Brian A. Jackson, conclude that technical measures alone are unlikely to change the underlying incentives that make distillation attractive, and argue that broader policy tools (beyond technical countermeasures) will likely be needed. Published September 2, 2026 as part of RAND's Expert Insights series.
Why it matters
For export-control and national-security policy teams, this reframes model distillation as a policy problem, not just an engineering one — informing how firms and agencies should think about IP protection and compute/access controls for frontier models.
Action needed
Brief export-control and IP-strategy teams on the limits of technical anti-distillation measures and the need for complementary policy safeguards.