Trump quote shows why an AI agency cannot work

The piece argues that the debate over AI regulation wrongly assumes governance starts only when government acts. It points to existing lab practices: DeepMind consulted more than thirty outside experts in biology, biosecurity, bioethics and human rights before releasing AlphaFold. Since January, California has required the largest labs to publish safety frameworks and report serious incidents within 15 days. Independent testers like Britain's AI Security Institute and METR now evaluate models before or around release.

The article's main warning is that these internal structures lack outside verification. It cites OpenAI's nonprofit board, designed as a check on its chief executive, which dissolved on contact with him. It also notes the EU's AI Act largely copied the regime firms had built, and argues that regulation follows governance rather than the reverse.

The proposed fix borrows from accounting. Standards were set by a private body in London in 1973, then severed from industry funding in 2001, with a monitoring board of regulators gaining a veto over appointments in 2009. Today more than 140 jurisdictions require those standards, and the US relies on its own private board, FASB. Applied to AI, such a board would lack a national owner, making its rules easier for other governments to adopt—and giving American firms a trusted passport to sell abroad.