What a framework is for
Public frameworks, such as the NIST AI Risk Management Framework and the ISO/IEC 42001 management system standard, are broad. They cover many kinds of AI and many risks. An operations team that needs to start this quarter often wants something smaller: a working loop for AI agents that act in the business.
The four steps
1. Know
Keep one list of every AI agent and tool in use, with a named owner, a purpose and the systems it reaches.
- Owner: operations lead, with IT.
- Output: the register.
- Check: can you name every agent running today?
82%
of surveyed enterprises reported AI agents running in their environment that they did not know about.
Source: Cloud Security Alliance survey, commissioned by a security vendor, April 2026. 418 IT and security professionals
2. Authorize
Decide what each agent may access, do and spend.
- Owner: the business owner of each agent, with risk.
- Output: a policy per agent: access, allowed actions, spend limit.
- Check: does any agent have more access than its job needs?
3. Operate
Run agents day to day with a person approving high-risk actions.
- Owner: named approvers and the operations lead.
- Output: approved or declined actions, an Emergency Stop plan.
- Check: who approves an email to a customer, and how fast?
4. Prove
Keep a record of every action, approval and cost, and review it.
- Owner: operations lead, with finance and compliance.
- Output: an audit log and a monthly review.
- Check: can you show last month's agent actions to a customer or auditor?
A monthly rhythm
| Week | Activity |
|---|---|
| 1 | Update the register: new, changed, retired agents |
| 2 | Review limits and approvals: too many or too few requests? |
| 3 | Review the record and costs: anything unexpected? |
| 4 | Report to leadership: what AI did, what it cost, what changed |
How this relates to public frameworks
The four steps are a working loop, not a replacement for public frameworks. They map loosely: Know and Authorize support what NIST calls Govern and Map, and Operate and Prove support Measure and Manage. This is an orientation, not a formal mapping. MFDIO is not certified to, audited against or endorsed by NIST or ISO. If your business needs to meet a specific rule or standard, work with a qualified adviser.
How MFDIO applies the framework today
The platform is built around these four steps: Know, Authorize, Operate, Prove.
- Today: register, policy controls, approvals, Work Units for spend, and an append-only audit log.
- Roadmap: agents built elsewhere in the register, live-data reports, SOC 2 and ISO 27001.
See what works today.
Questions people ask
Do we need a committee? Not to start. A named lead, a register and a monthly review are enough for a first version.
How long does it take to start? A first register and set of limits can often be drafted in days. The rhythm builds over a quarter.
Limits of this guide
This guide is general information and not legal advice. MFDIO is in invite-only early access and is pre-revenue. MFDIO does not hold SOC 2 or ISO 27001 today. Market figures are analyst or survey estimates and vary between sources.
Sources
- NIST, AI Risk Management Framework (AI RMF 1.0), January 2023: Govern, Map, Measure, Manage
- ISO/IEC 42001:2023, Artificial intelligence management system standard (referenced by name; not reproduced)
- Cloud Security Alliance, "Autonomous but Not Controlled: AI Agent Incidents Now Common in Enterprises", 21 April 2026 (418 IT and security professionals; commissioned by a security vendor)
- MFDIO product facts: checked against /status on 5 October 2026
Published . Last updated . Written by the MFDIO team.