What they are
The term comes from "shadow IT", where teams adopt tools without approval. Agents are easy to start: a person connects an AI to email, a spreadsheet or a customer system and lets it act. It often begins as a helpful experiment.
Why it matters
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
An unknown agent has no owner, no limits and no record. Nobody approves what it does, nobody sees what it costs, and nobody can say what it touched if a customer asks.
Why they appear
- A simple agent can be built in hours.
- Teams want results and do not want to wait for a process.
- There is no single place to register an agent, so nothing says it should.
- Employees use personal accounts or free plans.
Banning agents rarely works. A simple, fast path to register one works better.
How to find them
- Ask. Send every team lead one question: which automations and AI agents do you use, and what can each one do?
- Check connections. Review which tools and accounts have connected AI apps, and which service accounts and keys exist.
- Check spend. Look for AI charges on cards and invoices that no one owns.
- Check the logs of key systems for activity from automated accounts.
- Make registering easy. An agent that is registered gets help. An agent that is not gets reviewed.
What to record for each agent
| Field | Why |
|---|---|
| Name and purpose | What it is for |
| Named owner | Who answers for it |
| Systems it can reach | What could go wrong |
| Actions it may take | What needs approval |
| Spend limit | What it may cost |
| Review date | When someone looks again |
After you find them
Do not switch everything off. Sort agents by risk. Put a named owner on each. Add approval for anything that contacts customers or changes records. Set a spend limit. Retire what nobody uses.
How MFDIO applies this today
- Today: the agent register holds agents built in MFDIO, with the person who created each one on record, and an Emergency Stop for all agents. Approvals and spend limits apply to them.
- Roadmap: registering agents built elsewhere, assigning a named owner to each agent, and automatic discovery. Until then, finding outside agents is a manual process like the steps above.
See what works today.
Questions people ask
Can software find every agent automatically? Not reliably today. Combine asking, system checks and spend checks.
Are shadow agents always bad? No. They often show real needs. The risk is that nobody governs them.
Where do we start? Send the one question to team leads this week.
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
- 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)
- Gartner, June 2025: over 40% of agentic AI projects predicted to be canceled by the end of 2027, citing cost, unclear value and weak risk controls
- MFDIO product facts: checked against /platform/agent-register and /status on 5 October 2026
Published . Last updated . Written by the MFDIO team.