AI agent governance means knowing what your AI agents can access, what they're allowed to do, who's responsible when something goes wrong, and whether there's a record of what they did. Without that in place, an agent can take a significant action in a split second with nobody watching.

If your business is already using tools like Microsoft Copilot, Zapier with AI features, or an AI-powered customer service platform, you may already have AI agents running without a governance framework in place. This article explains exactly where to check for agents you may already be running, and the four minimum controls you need before deploying one.

If you are not sure whether your business is already using AI agents, the AI Readiness Assessment is the place to start.

How are AI agents different from the tools most businesses are already using?

Most Irish businesses that have adopted AI are using generative AI tools: ChatGPT, Microsoft Copilot, Google Gemini, and similar products. These tools respond to prompts. They produce output. A human reviews that output and decides what to do with it.

AI agents change that dynamic. Instead of producing output for a human to act on, agents act directly. They have access to systems, databases, and external tools. They can read data, write data, send communications, trigger workflows, and in some cases, make decisions that affect your customers, your finances, or your operations.

Trinity College Dublin's AI Economy Ireland 2026 report, produced with Microsoft Ireland, found that agentic AI is expected to have its highest impact in customer support, with supply chain management, knowledge management, and cybersecurity also identified as high-potential use cases.[3] Separate research from The Innovation Exchange in March 2026 found that around 29% of Irish SME leaders say they are already using AI agents, though that survey was commissioned by OpenAI, so treat it as a general indicator rather than a precise market figure.[4]

The distinction matters for governance because the human review step that exists when a person uses a generative AI tool disappears when you deploy an agent. The agent acts. The consequences follow.

In summary

With a generative AI tool, a human reviews the output before anything happens. With an AI agent, that review step disappears. The agent acts, and the consequences follow at machine speed.

Where should you check your agent settings before you do anything else?

Before you can control what your AI agents do, you need to know which ones already exist. For most Irish SMEs that means three places.

Start with Microsoft 365 if your business runs Copilot, Teams, SharePoint, or Outlook. The Microsoft 365 admin center has an Agent Registry that lists every agent in your tenant along with the channel it runs on: Copilot, Teams, Outlook, Microsoft 365 apps, or SharePoint. Use this as your starting inventory. From the same admin center, under Integrated Apps, you can enable, disable, block, or reassign any agent for the whole organisation rather than relying on individual staff to switch things off. SharePoint needs a separate check: the Microsoft 365 Copilot licence details page lets you toggle "Microsoft 365 Copilot for SharePoint" per user and edit the underlying service plan to allow or block Copilot experiences there specifically, which matters because SharePoint often holds your most sensitive file store. One default worth confirming rather than assuming: agents created in Copilot Studio and Teams are set to "Authenticate with Microsoft" by default, the safer setting. Check nobody has switched a customer facing or high access agent to run without it.[8]

If your business runs Zapier, task history and retention live under Settings, then Security and data, then Compliance and data. Zapier keeps Zap history for 30 days by default before deletion. Company and Enterprise plans can set a custom window from 7 to 30 days, or extend up to 90 days if you specifically need it for troubleshooting. If you are running any Zap with AI steps that touch customer or financial data, this is the setting that determines how long you can actually reconstruct what an agent did.[9]

If you are on n8n Cloud, execution log retention is fixed by plan: 7 days on Starter, 30 days on Pro, unlimited on Enterprise. If you are self hosting, retention is controlled by two settings: executions are pruned after 14 days by default, or once you pass 10,000 stored executions, whichever comes first, so check with whoever manages your instance that this has been set deliberately rather than left on the default. Separately, n8n can hide sensitive data from the workflow screen by default, only revealing it to specific people you grant access to, and instance owners and admins have that access automatically. If your n8n workflows process personal data, turning this on is worth doing regardless of your wider governance work.[10]

In summary

You cannot govern an agent you do not know exists. Before you write any policy, open the Microsoft 365 Agent Registry, Zapier's Compliance and data settings, and n8n's Admin Panel, because that is where your real inventory is.

What are the four minimum controls before deploying an AI agent?

These four controls do not require a large IT team or a compliance function. They are practical decisions any business owner or operations manager can implement.

1. An agent inventory. Before anything else, document every AI agent currently running in your business. This includes agents embedded in software platforms you already use (the section above sets out exactly where to look in Microsoft 365, Zapier, and n8n), agents built by staff using no-code tools, and any AI-powered automation that takes action without human approval at each step. If you cannot produce this list, you do not have governance. Review the inventory at least quarterly. SaaS platforms add agent capabilities through routine product updates, and staff build new shadow agents continuously, so a one-time inventory goes stale fast.

2. Minimal access privileges. Each agent should have access only to the specific data and systems it needs to complete its defined task. Not your full CRM. Not your entire file store. The specific fields, folders, or records the task requires. This limits the potential damage if an agent malfunctions or acts on incorrect information.

3. Human approval for high-stakes actions. Define which types of actions require a human to confirm before the agent executes them. A reasonable starting point for most Irish SMEs includes any action that involves sending external communications, processing financial transactions, modifying customer records, or deleting data. A useful rule of thumb for anything not on that list: if reversing the action would take more than an hour of manual work, or if it would visibly affect a specific individual, gate it behind human approval. The agent can prepare the action. A human confirms it runs. This gate does more than manage operational risk. Where an agent's action produces a legal or similarly significant effect on a person, confirming a price, declining a service, finalising a booking, with no human involved, GDPR Article 22 restricts fully automated decisions of that kind. Keeping a human in the loop on high-stakes actions is what keeps most everyday agent use outside that rule.

4. A retained audit log. Every action an agent takes should be logged with a timestamp and retained. The section above sets out where to find and configure retention in Microsoft 365, Zapier, and n8n specifically, and what each one defaults to. Turning on the native log is not enough on its own if it is left to purge automatically before you would ever need it. This serves two purposes. First, it allows you to reconstruct what happened when something goes wrong; if an agent does take an unintended action involving personal data, treat it as you would any other data incident and start your GDPR breach assessment process immediately, not once you have finished investigating internally. Second, if you are deploying a high-risk AI system, Article 26(6) of the EU AI Act requires you to keep the logs it generates automatically for at least six months.[1] That requirement was originally due to apply from 2 August 2026, but under the Digital Omnibus agreement from May 2026, it is now expected to apply from 2 December 2027 instead. The agreement still needs to be formally published before that date is locked in, but if your agent falls under a high-risk Annex III category, keeping that log will not be optional once the rules take effect.

Before any of these four controls matter, one prerequisite applies regardless of risk classification. If an agent will process personal data, customer, employee, or supplier data, through a third-party platform, that platform provider must have a Data Processing Agreement in place under GDPR Article 28. This is not a best practice you can defer. It is a legal requirement from the moment personal data changes hands, and the Irish Data Protection Commission's guidance on AI and data protection sets out the baseline expectation.[5] Confirm the DPA is signed before you connect an agent to any system holding personal data, not after.

These four controls are the foundation of an agent governance framework for an Irish SME, and a base you can build on from there. For what a complete governance and workflow structure looks like once these foundations are in place, see AI governance and workflow blueprint: what it is and when an SME needs one.

In summary

The four minimum controls for any AI agent are a documented inventory, minimal access, human approval gates for high-stakes actions, and a retained audit log. Put those in place and you have the foundation to build the rest on.

Does the EU AI Act apply to AI agents in Irish businesses?

How much regulatory risk your AI agent carries depends entirely on what it is used for, not on the technology itself. An agent that drafts your marketing copy is treated completely differently to one that helps decide who gets a loan, who gets hired, or what someone pays for insurance, because those decisions can seriously affect a person's life. Article 6 of the EU AI Act sets out two ways a system can be classed as high-risk. For most Irish SMEs, the one that matters is Annex III, which covers AI used in areas like employment, credit, biometrics, and law enforcement. The European Commission published draft guidelines on 19 May 2026 clarifying how this classification works in practice: an agent built from several components, for example a workflow that combines a general-purpose model with a scoring tool and a CRM connector, is assessed as a whole. A single high-risk component can pull the entire configuration into the high-risk regime even if the other parts would not qualify on their own. The guidelines are still in draft, but they show the direction regulators are heading.[6]

If your agent falls into a high-risk category under Annex III of the EU AI Act, Article 26 applies to you as the deployer. Once those rules take effect, you will need to put human oversight in place, monitor how the system is performing, keep logs for at least six months, and, under Article 26(7), tell affected workers and their representatives before the system goes live in the workplace.[1] There is a separate rule too: AI literacy under Article 4 applies to everyone using or providing any AI system, whatever the risk level, and it has been in force since 2 February 2025.[2] The high-risk obligations under Article 26 were originally due to start on 2 August 2026. After the Digital Omnibus political agreement in May 2026, they are now expected to start on 2 December 2027 instead for stand-alone Annex III systems. That change is not yet formally published, so the old dates are technically still law, but December 2027 is the date to plan around.

Most AI agents used in a typical Irish SME today are not high-risk under the current classification. An AI agent that influences recruitment or employee evaluation decisions would fall under Annex III category 4 (employment and worker management), and one that influences creditworthiness decisions would fall under category 5(b) (access to essential services, including credit). Either means the full set of Article 26 obligations would apply once those rules take effect. But the minimum controls described above are sound business practice regardless of your regulatory position. They protect you operationally before they protect you legally.

If you are uncertain whether a specific agent or AI system your business uses falls into a high-risk category, the AI Readiness Assessment includes a risk classification assessment as part of its scope.

In summary

Most AI agents used by Irish SMEs today are not high-risk under the current classification. The minimum controls in this article are sound business practice regardless of regulatory position. They protect you operationally before they protect you legally.

If you want to build a governance framework before deploying AI agents in your business, the AI Policy and Governance Pack gives you the structure you need.