Governance & Operations

AI Governance Starter

AI policies and an inventory of every model and agent, kept current by a check in CI.

GIST stepGroundTrust
Download the PDF

Get the AI Governance Starter overview

A short PDF to share with your team. It covers what is in the package, how it works, and a worked example.

The problem

The board asks how you govern AI. The answer is scattered.

AI is already in use through vendor features, pilots, models, and agents. The policies are drafts, the inventory is a spreadsheet nobody updates, and each customer security questionnaire starts from a blank page.

What's included

Governance records that stay current.

01

Policy and acceptable use

A governance policy and an acceptable-use standard for staff, for your counsel to review.

02

Risk questionnaire and inventory

Scoring, with the controls each risk tier requires, and an inventory of models, vendor AI features, and agents.

03

Evaluation Cards and Control Register

A card for every live system and a register of starting controls, each with an owner and evidence.

04

Evidence packs

Built on request for auditors, customer security questionnaires, and board reports.

How it works

Setting it up.

  1. Records folder

    The inventory, control register, assessments, and cards live in your repository and are reviewed like code.

  2. Adapt the policies

    Set roles, approved tools, and prohibited uses to match your position and the law where you operate.

  3. Tune the questionnaire

    Scores, thresholds, and controls per tier, approved by your AI governance committee.

  4. Fill the register

    An owner, an enforcement method, evidence, and an honest status for each control.

  5. Register what exists

    Every AI system in use today, assessed, with its gaps listed.

  6. Check in CI

    A change fails when an inventory entry is missing fields or overdue for review, a live system lacks a current assessment or card, an agent’s scope file does not match the inventory, or a control its tier requires is not in place.

Worked example

Three AI systems, fully documented.

A demand forecast model, a support-answers assistant, and an invoice-triage agent, each assessed and given an Evaluation Card.

One command builds an evidence pack from
  • The AI inventory and each system’s risk tier
  • Risk assessments and Evaluation Cards
  • The Control Register, with owners and evidence
  • Test results and audit logs from the other accelerators
Result

The pack can cover every system or only the ones a given customer uses. A coverage gap on a production system fails the check.

Where it fits

What it runs on, and where it stops.

Works with

  • Your version control and existing CI
  • Plain CSV, YAML, and Markdown that open in Excel
  • GRC tools through CSV import

Assumptions and limits

  • Policies are starting drafts and are not legal advice. Your counsel reviews them
  • The crosswalk to NIST AI RMF and ISO/IEC 42001 is informal, and your auditor checks it
Ask Dex

Ask about the AI Governance Starter

Dex answers from our own pages and links to where it found the answer.

Is a security questionnaire sitting in your inbox?

Send us the questions. We will show you which ones an evidence pack would answer.