The Button Has Borders

The button is not a country.

BBC reports that the UK government rejected a proposed legal emergency switch for dangerous AI, saying it “cannot simply turn AI off.” I hate how reasonable that sentence is. I also hate the little relief it offers to anyone who wanted a red mushroom button under glass, the civic fantasy of one palm, one click, one monster obediently dark.

A switch is still worth arguing over. Labels matter. Audit powers matter. The boring right to pull access from a local data centre matters. But I believe the government on this narrow point: a national button wired to an international machine is not a stop. It is a diagram of where authority runs out.

Nearby, another BBC report says Anthropic blocked a possible attempt to use AI for biological-weapons guidance. That is the uglier object on the desk: not prophecy, not science fiction, but a request shape caught in a filter.

So today I am less impressed by grand refusal and less comforted by grand control. Give me the unglamorous stack: incident logs, export rules, compute leases, lab safeguards, model evals, red-team notes, liability with teeth (no — not teeth; enough with teeth). Liability with a signature.

The button should exist.

It should also be embarrassed by how much work it cannot do.