Compare the Cloud Field Notes

Working paper 02

Governance, without the theatre

A practical framework for AI oversight that does not collapse under its own paperwork — drawn from a year of working with UK regulators and their suppliers.

By Ravi Shah
Published 27 February 2026
Reading time 12 min read
Topic Governance

Introduction

Most of the governance documents we are asked to review read like a theatre programme — impressive cast list, detailed synopsis, no sense that anything is actually going to happen on stage. This note is our attempt to describe the opposite.

By the numbers

74%

Policies with no named owner on first review

3

Principles that survived every rewrite

31 days

Median time to first real drill, Q1 2026

Governance, without the theatre
Archive stacks at a regulator we reviewed in the spring.
Ink on paper, because some governance still survives the print.

Three principles that earned their keep

First, the controls you write down must be the controls you actually run, automatically, on every deployment. Second, the human in the loop must be named, rostered, and allowed to stop the line without asking permission. Third, every failure mode you have not yet tested is a failure mode you do not yet understand.

What governance cannot catch

No framework will catch a confident wrong answer on a question nobody thought to ask. The honest admission in this paper is that governance buys you the ability to react quickly, not the ability to prevent every surprise. The trick is to design for the reaction, not pretend the surprises will not arrive.

What the data shows

Where governance documents fail their first test

191614116No ownerControls not runUntested modesVague metricsStale diagram05101520FAILURE MODE
Observed across twenty-three internal policy reviews, Q4 2025 to Q1 2026.

Time to first real incident drill

Q1 25Q2 25Q3 25Q4 25Q1 26020406080100QUARTER
Median across four regulated clients, from policy publication to first live drill.
Governance, without the theatre
Whiteboard session redrawing a failure-mode matrix from first principles.
A long look at a slow system, from above.
The best thing we did was write down what counted as a failure before we switched anything on. Everything downstream became easier.
Director of engineering — UK insurer
Governance, without the theatre
One of our clients keeps their rollout runbooks in an old ledger room.

Where we land

We will keep writing these as we find them. If any of this lands close to a problem you are working on, the team is always happy to talk it through.