AI Defaults Are Unsigned Policy

When you deploy any AI system, someone sets the parameters that determine how it will behave: which minimum score triggers an action, which risk level is considered acceptable, what gets escalated for human review, and what passes straight through.

These are technical decisions. But in practice, they function as corporate policy. They govern more decisions per day than any document ever approved in a board meeting.

The problem is that they were made by the systems architect, the vendor, or the most technical person on the project - within the deadline, under delivery pressure, without formal approval, and with no record of the alternatives that were considered and discarded.

This is the blind spot of AI governance at most companies.

All the focus goes to acceptable use policies, privacy terms, and data protection compliance. The parameters that define how the AI decides remain untouched since go-live.

A score threshold calibrated for a customer profile that has already changed. A risk threshold defined based on data from last year's project. A default behavior no one questions because 'it's always been this way since launch.'

Your company reviews contracts every year. It adjusts targets every quarter. It updates the org chart when the business changes.

When was the last time you reviewed the parameters that define how the AI decides?

Tell me in the comments: is there an AI system running in your company whose decision defaults you don't know who set, when, or why?