The Private Capture of Public Genius

Cameron Russell Armstrong argues that frontier AI labs are capturing public knowledge into private model weights without compensation — a replay, in his telling, of the bargain that built Silicon Valley, but without the public benefit that came with it.

The essay opens with AT&T's 1956 consent decree, which forced the company to license all 7,820 unexpired patents royalty-free. Because AT&T's profits were capped at a regulated return on its capital base, Bell Labs research was both a recoverable cost and a rate-base expander — a self-reinforcing cycle that produced the transistor, the solar cell, and information theory. Gordon Moore later called the decree one of the most important developments for the commercial semiconductor industry. Armstrong's point: the public got something back.

Today's frontier labs, he contends, give nothing back. They train on "the collective experience, knowledge and learnings of humanity" (quoting Sam Altman) and compress it into private model weights. Anthropic's annualized revenue grew from $87M to $47B in just over two years — what Armstrong calls "the fastest-compounding enterprise software company in history."

The essay surveys the legal landscape: Bartz v. Anthropic (training on legally acquired books is transformative; training on pirated books is not), Kadrey v. Meta (LLM training transformative, but plaintiffs failed to prove market harm), and a 2025 U.S. Copyright Office report concluding that public availability does not inherently permit fair use training.

The core problem is attribution. Labs claim individual training inputs are "worthless" because removing any single one barely affects the model. Shapley value — the leading method for valuing contributions — is computationally infeasible at frontier scale. Armstrong concludes that payment is still owed, just not proportionally.

His policy proposal: a corpus royalty — frontier labs pay a fixed share of gross revenue into a public fund distributed equally to every eligible American. He models it on the Alaska Permanent Fund (distributing resource wealth no individual can separately claim) and Superfund (holding polluters accountable without tracing each barrel). The royalty is framed as restitution, not welfare or a tax.

The essay closes by drawing on Elinor Ostrom's conditions for sustainable commons governance, noting the internet satisfies almost none of them. Armstrong warns that frontier labs seek "the privilege and power of a utility without accepting the public obligations that come along with it."