Two things happened in AI governance this week that, read together, describe a complete system. Trump signed a revised AI executive order that swapped mandatory oversight for voluntary prerelease government reviews, after industry objected to the original. The same day, OpenAI launched new Codex tools aimed squarely at white-collar enterprise workflows. The sequence is not a coincidence. It is the product.

When Regulation Becomes a Feature, Not a Constraint

The narrowing of the executive order is a case study in what Hayek might have called the worst getting on top. The Atlantic's piece this week on Hayek and institutional capture is almost too on-the-nose: voluntary compliance lets the most aggressive players set the norm, while the cautious ones get punished for caution. OpenAI's Codex rollout illustrates the flip side. When state oversight is optional, the oversight infrastructure gets absorbed into enterprise SaaS. Compliance becomes a selling point on a pricing page, not a public good. A 2023 paper in AI and Society by Jobin et al. found that voluntary AI ethics frameworks consistently converge on principles but diverge on enforcement, producing what the authors called "ethics washing." The 2026 executive order is ethics washing with a presidential seal.

The White-Collar Automation Anxiety Underneath

Codex is being positioned for "white-collar work," which is the industry's way of saying: the creative class is next. The New Yorker's Jay Caspian Kang asked this week whether AI can produce writing we actually want to read, concluding it cannot. But "want to read" and "will pay for" are different markets. Enterprises will pay for writing that is good enough, consistent enough, and cheap enough. Codex isn't trying to win a Pulitzer. It is trying to make knowledge work as frictionless as a SaaS invoice. The voluntary oversight regime isn't a bug in this picture. It is the subsidy.