In 2020, an automated license-plate reader in Aurora, Colorado matched a plate number to a stolen vehicle report — but not the vehicle’s state or type. A patrol car pulled the vehicle over at gunpoint. At 87 degrees, they held the family for nearly 2 hours.
It was the wrong vehicle.
How is that possible?
Nobody had decided, in advance, that a partial match wasn’t enough to act on.
AI is running. Nobody designed what the human does next.
That gap — between machine output and the person who has to act on it — is where most AI deployments quietly fail. Not because the technology is wrong. Because nobody asked the right question before deployment: who is this decision for, and what are they authorised to do when the AI is confident and wrong?
Think of it as a fence. Someone has to mark the boundary before it can be built or enforced. That boundary is the spine — it decides what’s safe to automate and what isn’t. Most organisations build the fence — the guardrails — without ever having built the spine.
You can’t decide in the moment — there’s no time to say no. The decision happens before it runs, or it doesn’t happen at all.












