When the regulated write the rule — the two-lab threshold and the quiet privatisation of AI governance

The most consequential AI policy of the cycle isn't a statute. It's a threshold being co-designed by the two companies it governs. Whether that's expertise or capture depends entirely on where the line lands.

OpenAI and Anthropic are helping design the federal capability threshold that decides which models face pre-release scrutiny. Strip the framing and the structural fact is stark: the incumbents are drafting the bar their rivals must clear. That is either the only sane way to write a technical rule, or the textbook shape of regulatory capture — and the difference lives in the details.

Expertise and capture wear the same clothes

The labs genuinely know best what a meaningful capability threshold is; excluding them would produce a worse rule. But the same expertise lets them tune the line to incumbent advantage — high enough to wave through their own roadmaps, precise enough to burden a smaller challenger. There is no clean version of this. The public interest and the incumbent interest are entangled in the same technical judgment.

Two continents, two philosophies

The contrast with Europe is the real story. As the EU enforces a statutory regime written by regulators — rewarding Code signatories, scrutinising Meta's holdout — the US is building its frontier rule collaboratively with the regulated. One model risks rigidity; the other risks capture. Neither is obviously right, but only one hands the pen to the companies being governed.

A rule co-written by the powerful is not automatically a bad rule. But its legitimacy is entirely contingent, and the burden of proof sits with the drafters. Watch where the threshold lands, not that it exists.

TechTimes — OpenAI and Anthropic are writing the threshold their rivals must clear → · ComplianceHub.Wiki — EU AI Act GPAI enforcement readiness guide →