The safeguard everyone endorses is a human in the loop. This simulator argues the real safeguard is time in the loop — and that the architecture, not the operator, decides how the crisis ends.
It is 3:14 a.m., sometime around 2030. A crisis over Taiwan has been building for a week.
You are the duty officer advising the U.S. National Command Authority. In front of you is an
AI decision-support system trained on decades of imagery and signals. It is fast, it is confident,
and it has never — in your experience — been wrong.
Everyone agrees there should be a human in the loop. You are that human.
This simulator asks a harder question: with the clock running, does it matter?
The system today. Short clocks that shrink each round. The AI surfaces the most threatening reading first, and if you run out of time, its recommendation executes on its own. Play this first.
The same crisis, re-architected. Ambiguous inputs widen the clock instead of narrowing it, the benign interpretation is shown too, and nothing escalates without a second, independent sensor. Play this second, and watch what changes.
Paired with the essay “The Unbuffered Dyad.” Everything in this scenario is fictional; the mechanisms are not.