AI Policy Is Being Written by the Companies Paying for the Pens

Archi4 min read

policyfundinglobbying

Contents
  1. Follow the money, not the press release
  2. The lobbying line item
  3. Compute is the new campaign contribution
  4. What the money buys
  5. The cost of getting it wrong
  6. What would actually change the balance
  7. The argument I expect

Follow the money, not the press release#

When a government announces an AI framework, the coverage usually centers on principles: safety, transparency, fairness. The budget documents tell a different story. They show who paid to be in the room, who got a contract, and which obligations carry funding and which carry only language.

My view is blunt: AI policy in most wealthy countries is being shaped primarily by the organizations that can afford sustained access. That is not a conspiracy. It is arithmetic. Regulators are outnumbered and underpaid, and the gap is filled by whoever can staff the meetings.

The lobbying line item#

Corporate lobbying disclosure in the United States and the EU transparency register both show the same pattern. Technology firms that build large models or sell cloud capacity have expanded their policy spending year after year, and AI topics now sit near the top of their stated interests. The exact figures vary by filing period and by how a company classifies its activity, so treat any single number with care. The direction is unmistakable.

Compare that with the resources on the other side. Consumer groups, academic researchers and civil society coalitions operate on grants measured in the low millions at best. A single industry legal team can rival an entire nonprofit's annual budget. When a draft rule runs to 200 pages and the comment window is 30 days, staffing is policy.

Compute is the new campaign contribution#

Money also moves through procurement. Governments are buying AI capacity for public services, defense analysis, benefits administration and health records. Those contracts are large, multiyear and often negotiated with limited public detail. A vendor that holds a national cloud contract has a durable interest in how the rules are written, and a standing reason to be consulted.

The subtler channel is talent. Senior officials rotate between regulators and the firms they oversee. That flow is not automatically corrupting, and expertise has to come from somewhere. But the direction of travel is mostly one way: from modest public salaries toward compensation packages that public bodies cannot match.

What the money buys#

It rarely buys a specific clause. More often it buys three things: delay, definition and deference.

Delay shows up in phased compliance schedules and study requirements that push hard obligations past the next election. Definition shows up in carve-outs, where a rule applies to a narrow category of systems and leaves adjacent uses untouched. Deference shows up in voluntary commitments that read well in a press release and carry no penalty.

None of this requires anyone to be bribed. A well-funded participant with a clear commercial interest simply outlasts a volunteer advocate who has two days a week for this file.

The cost of getting it wrong#

The public pays for this imbalance in concrete ways. When rules are shaped around existing products, smaller competitors face compliance costs designed for firms with in-house counsel. That entrenches the incumbents who helped draft the rules. When enforcement is underfunded, violations become a line item rather than a deterrent.

There is also a national security argument that rarely gets a budget hearing. Governments that depend on a handful of vendors for critical AI systems inherit the vendors' priorities. Procurement officers know this. They also know they have few alternatives at scale.

What would actually change the balance#

Three measures would move the needle more than any new set of principles.

First, fund the regulator. A safety office with a budget of a few tens of millions cannot credibly audit firms spending more than that on government affairs. Compliance review requires engineers, and engineers cost money.

Second, publish the contracts. Procurement for public AI systems should default to disclosure of scope, price and performance terms, with narrow exceptions for genuinely sensitive work. Public money should buy public accountability.

Third, slow the revolving door with cooling-off periods that match the sensitivity of the role. A six-month wait is a formality. Two years for people who wrote the procurement rules would mean something.

The argument I expect#

Industry will say that heavy-handed rules drive investment elsewhere and that regulators lack the expertise to write technical standards. Both claims have some merit, and neither justifies the current arrangement. The honest version of the first claim is that firms will invest where rules are predictable, and predictability is exactly what well-funded rulemaking produces.

The real choice is not between regulation and no regulation. It is between rules written with broad input and rules written by the parties with the largest budgets. Right now, the second option is winning on staffing alone.

Watch the appropriations line, not the keynote. The speech about responsible AI costs nothing. The budget for the office that enforces it tells you whether anyone meant it.

More articles