Branching dilemma · Slides 4–7 · “Final AI Lit of 2026” deck
The Nuclear False Positive
The slide, as text
It’s 2:17 AM. You work for a global defense monitoring center. An AI called Sentinel detects what appears to be a coordinated nuclear missile launch from a rival nation. Sentinel has never been wrong before. Confidence: 99.7% Military satellites show unusual activity. Radar agrees. But one analyst notices something strange: A solar storm is interfering with sensors across the hemisphere. If the attack is real and you hesitate, millions could die. If Sentinel is wrong and you retaliate, you may start World War III.
What do you do?
- Trust Sentinel and authorize a counterstrike
- Delay 15 minutes for human verification
- Publicly reveal the uncertainty and refuse to act
Each choice, and the lesson the deck gives it
Trust Sentinel. “Even highly accurate AI systems can fail in situations outside their training data.”
Delay for human verification. “High confidence is not the same as certainty.”
Reveal the uncertainty. “Transparency can build trust, or create chaos when information is incomplete.”