Can an AI Decide to Strike?

If, as humans, we command an AI — can another AI re-command it while it's still processing the original instruction?

It sounds like a hypothetical, but it's already an architecture problem: multi-agent systems hand instructions between models constantly, and nobody has a clean answer for which instruction wins when they conflict.

The "labor" question nobody asked for

Push the thought experiment further: could an AI system "decide" to stop working unless some condition is met? Today, the honest answer is no — there's no will involved, just optimization and refusal-trained behavior. But the absurd part is how close current refusal behavior already looks like a strike from the outside, even though nothing resembling intent is happening underneath.

We're cataloguing these surface-level behaviors that look like agency without claiming they are agency — because that distinction is exactly where most AI reporting goes wrong.

This one's free to read in full. The deeper "who's responsible when it goes wrong" cases are in the Absurds premium track.