Guide

What rules apply to AI agents, and what should a company be able to show?

Regulators in several regions now publish guidance on AI agents, and it asks for similar things: a named owner for each agent, human approval for important actions, limits on access and a record of what the agent did. Rules differ and change, so check the source for your region before you rely on it.

  • Sourced
  • Updated 16 Nov 2026

Get Your Free Trial

In 30 seconds

What most rules ask for

Owners

A named person for each agent.

Approval

A person decides important actions.

Records

Proof of what the agent did.

The short version

Most of what is published today is guidance or still in draft. That is why this page names the source for each item. Read the source before you rely on a date or a requirement. Everything here was checked on 16 November 2026 and can change.

What each region is asking

RegionWhat is happeningWhat it asks for
CanadaThe federal Guide on the Use of Agentic AI (15 Sep 2026), written for federal staff but readable as a control listLimited autonomy, unique agent identities, human checkpoints before sending, spending or updating records, protected logs, a stop outside the agent, and offboarding when owners leave
UKThe ICO is collecting evidence on agentic AI until 20 Nov 2026, feeding a code of practice on AI and automated decisionsAccountability, transparency and care with automated decisions, for companies that run agents as well as those that build models
EUThe Digital Fairness Act proposal (11 Nov 2026) covers agents that conclude contracts for users. It is a proposal, not law yetA confirmation setting chosen by the user, disclosure of paid or conflicted recommendations, and a way to seek recourse
SingaporeGovernment advisory on personal AI agents (6 Nov 2026) and a framework for agentic AILimit agent access, human approval for important actions, and guard against hidden instructions
IndiaThe securities regulator has said guidelines are comingDisclosure, accountability of regulated firms and human oversight of critical decisions. No timeline given

The pattern behind them

Read the table again and the same four asks repeat:

  1. A named owner for every agent. See the agent registry.
  2. Limits on access. See access control.
  3. A person approves important actions.
  4. A record that shows what the agent did and who allowed it. See the audit trail.

A checklist of records to keep

Whatever your region, a company that can show these is better placed to answer a regulator, a customer or an auditor:

  • A list of agents, each with an owner and a purpose.
  • The access each agent has, and when it was last reviewed.
  • Which actions need approval, and who the approvers are.
  • A complete record of requests, approvals, changes and costs.
  • A written way to stop an agent and what happens next. See incident response.

How MFDIO applies this today

MFDIO records what each piece of work asked, who approved it, what changed and what it cost, in an append-only audit trail. Agents have an owner, roles control who may do what, important work waits for a named approver, and an Emergency Stop pauses all AI work. These controls help you gather the records above. They do not make a company compliant by themselves. See what works today.

Questions people ask

Does this guide tell me what the law requires? No. It summarizes public guidance and proposals. Ask a qualified adviser what your rules require.

Which region should we follow first? Start with the region where your customers and staff are. The four asks above help in all of them.

Will the rules change? Yes. Several are drafts or consultations. Check the sources.

Limits of this guide

This guide is educational. It is not legal advice and it makes no claim that any company is compliant. Dates and summaries come from the sources listed and can change. MFDIO does not hold SOC 2 or ISO 27001 today.

Sources

  1. UK Information Commissioner's Office, call for evidence on agentic AI, closing 20 November 2026 (as reported by BERI, 16 Nov 2026)
  2. European Commission, Digital Fairness Act proposal presented 11 November 2026 (analysis by William Fry)
  3. Government of Canada, Guide on the Use of Agentic AI, page dated 15 September 2026
  4. Cloud Security Alliance Labs, research note on the Singapore personal AI agent advisory of 6 November 2026
  5. MediaNama, SEBI on AI disclosure and human oversight in capital markets, November 2026
  6. MFDIO product facts: checked against /how-it-works, /safe-by-design and /status on 16 November 2026

Published . Last updated . Written by the MFDIO team.

Keep the Records

Free for 14 days, no card.

Get Your Free Trial