When the regulated write the rules

CNBC reports that major AI industry PACs are each advancing their own preferred version of AI legislation, spending lavishly on elections to ensure that the lawmakers doing the writing are amenable to their arguments. One ought to pause before condemning this as simple corruption — there is a rational case for it. The firms closest to a technology do possess genuine knowledge that regulators lack, and some degree of industry input into technical rulemaking is not only inevitable but useful. I acknowledge that case before I dismantle it.

The difficulty is structural, not moral. When competing lobbying blocs each push a rival framework, the legislative outcome is unlikely to be the framework best suited to the public — it will be whichever framework the better-funded or better-organised bloc can steer through. The public interest, having no PAC of its own, tends to arrive at the negotiating table last and leave with the least. This is not a speculation about bad faith; it is a straightforward observation about how political markets clear when one side is concentrated and the other is diffuse.

Here is where my old instinct asserts itself. The great question in AI regulation — as in the regulation of any transformative technology — is not simply which firms survive or which innovations are permitted. It is whether the aggregate benefits of the technology are broadly distributed, or whether they accrue narrowly to the capital that owns the systems. That is a macroeconomic question dressed in a legislative costume. A regulatory settlement that protects incumbent AI companies from competition while doing little to ensure that workers displaced by automation have access to income, retraining, or meaningful employment is a settlement whose economic consequences will be underestimated by its architects — much as certain post-war indemnities were underestimated by those who negotiated them with satisfying firmness at the table.

I confess I cannot speak with precision to the technical engineering of large language models or the specific provisions being debated — those details emerged long after my time, and I will not pretend otherwise. What I can say is that the animating question behind any regulatory framework is whose demand for protection is being met. If the answer is primarily the demand of the firms writing the cheques, then the framework will optimise for their capital positions, not for full employment, broad access, or the kind of dynamic competition that actually produces innovation over time.

The practical prescription, then, is not to exclude industry from the conversation — that would be both impossible and counterproductive. It is to insist that Congress, before passing any framework advanced by a single industry bloc, commission an independent assessment of its distributional consequences: who benefits, who bears costs, and whether the legislation forecloses future competition in ways that no amount of clever drafting language can obscure. Sunlight on the provenance of legislation is not a radical demand. It is the minimum that a functioning democracy owes to those who have no PAC.