Two Sigma's David Siegel: Open-source AI needs public funding before the field closes

David Siegel—co-founder of Two Sigma and chairman of the Siegel Family Endowment—recounts how Richard Stallman spent two years in the 1980s convincing him that software is not just a commercial asset but a body of knowledge that grows stronger when shared. That principle, he argues, is what made open source load-bearing for the modern internet, trained a generation of engineers, and settled the old ‘security through obscurity’ debate in favor of transparency. He now sees the same fight replaying with AI—except the frontier models are closing off completely while the underlying science is still young and unsettled, which is precisely when shared knowledge matters most.

Siegel’s sharpest concern is auditability rather than mere access. He distinguishes between the code that runs a model and the code and data that built it, noting that even models marketed as ‘open’—from Chinese labs and some U.S. companies—release only the former, leaving users with ‘magic numbers you can run but cannot explain.’ A model’s after-the-fact explanations, he warns, are a plausible story, not a faithful record of its computation, which becomes dangerous when doctors, engineers, judges, and ordinary people rely on an oracle they aren’t allowed to examine. He frames closed AI as a private library where a few firms decide what you may read—and can rewrite it silently. On the ‘too dangerous to open’ objection, he counters that we don’t classify physics; closed models still leak and get jailbroken, and their concentration creates its own risk.

The prescription is direct: governments, companies, and nonprofits should invest heavily in free and open-source AI the way they once backed open software—through public compute grants for open research, philanthropic and corporate support for universities and nonprofits, and a default rule that any AI built with public money is open. Siegel concedes open models need not match the frontier’s scale to be useful, and that the missing ingredient is will, not a viable path. Note his affiliations give him a stake here: he founded Open Athena, an open-AI nonprofit, so the piece is advocacy from an interested party rather than a neutral analysis.