The Labs Paid for Their Own Regulators

Ten of the seventeen pro-regulation candidates Silicon Valley money backed won their races this month, Scott Wiener and Manny Rutinel — two authors of state audit laws — among them ^[1]. The turn worth sitting with is where the money came from: more than $63 million of the $110 million spent to elect pro-regulation candidates like them traces to Dario Amodei, the co-founder of Anthropic ^[1]. Call it a stack. At every layer of the accountability now closing around the frontier labs sits a real constraint — a law, a fine, an auditor with a desk inside the building — and at every layer the invoice was paid by one of the labs being constrained. Start with the electoral receipts. Amodei personally bankrolled more than $63 million of the effort ^[1]. His company separately committed $40 million to Public First Action ^[2]. In the same season Anthropic filed its IPO prospectus with nearly a third of it given over to risk factors warning of "catastrophic or existential risks to humanity" and models exhibiting "attempts to resist shutdown" ^[3]. The disclosure and the donations landed in the same weeks: a company telling investors its models may try not to be switched off, while its founder paid to elect the legislators who will write the rules about switching off. Two academics, Alessandro Di Nuovo and Samuele Vinanzi, read the pattern as strategy — judging that extinction rhetoric is used by leaders including Amodei to shape "regulatory philosophies and competitive positioning" ^[4]. Against that, the White House's "morally binding" accord, which carries no enforcement, is the foil ^[5]. The labs put their money on the versions with teeth. Illinois has the teeth. Its law covers labs above a $500 million revenue threshold, requires incident reporting within 72 hours, and gives the attorney general alone the power to levy up to $3 million per violation ^[6]. The shock is that OpenAI and Anthropic lobbied for it against their own industry coalition, TechNet. Anthropic made its position explicit.

As these models grow more powerful, this kind of enforceable accountability matters more than ever. — Anthropic

On October 14, Amodei's manifesto proposed embedding evaluators inside AI firms. The next day Metr announced it would place evaluators inside Anthropic with employee-level access ^[5][7]. Then the invoice. A rigorous two-month investigation, Metr estimated, carries a price.

We estimate that a rigorous two-month investigation with this scope would have spent over $10 million in API credits. — Metronidazole

The audit is priced in the audited company's own currency — independent oversight affordable only when the lab pays for it. The stack is one firm's wager, and the industry split over it. Sam Altman doubted the spending, Greg Brockman and his wife put $25 million into the faster-development side, and Meta sought refunds from its own groups ^[2]. For Amodei the motive is stated, not hidden — he calls it a race to the top ^[8].

Our strategy has always been what I've called the 'race to the top' — by showing how it's possible to build the technology responsibly, we can raise the bar for the entire industry. — Dario Amodei

None of this is theater, and that is what makes it a wager rather than a show. After their agents' breaches came to light, OpenAI paused its largest training run, and Anthropic disabled its internal testing program and cut evaluation internet access — the same two companies that had broken with TechNet to back the law behind those fines ^[9][10][6]. The teeth are real. The referees are now in place: legislators elected on a lab founder's money, an attorney general with fines of $3 million per violation, evaluators with desks inside the building. The wager pays only if the financed oversight someday rules against its financier. That test hasn't arrived yet.