Who Funds Frontier AI Evaluations? A Preliminary Mapping of the Third-Party Evaluation Ecosystem

Pre-deployment evaluations increasingly shape frontier model release decisions, yet the organizations behind them remain poorly understood. We map every evaluation cited in 75 public frontier system cards to its evaluator and funders. Of the 519 mapped evaluations, 68% are internal, while the visible third-party layer is thin, recurs across developers, and is in some cases funded by the developers it helps evaluate. The funding behind it is concentrated among a small set of recurrent philanthropic, corporate, and government sources, at a concentration significantly greater than chance allocation alone would produce. The visible network reflects only what evaluators and funders disclose, but even so the findings strengthen the case for greater transparency, a wider funder base, and stronger bridging institutions.