Training: Secure and Govern your AI agent workforce (AB-6007)

When AI agents become a fleet to administer

Organisations started with one agent, then ten, then dozens spread across teams. Each of them reaches data, triggers actions and talks to users. The question is no longer whether AI enters the company, but who approves those agents, with which permissions and for how long.

An agent left without an owner raises the same issues as a forgotten service account: excessive privileges, invisible activity, decommissioning that never happens. Microsoft released AB-6007 on 28 May 2026 to answer that operational reality, in a deliberately dense half day format.

What Microsoft Agent 365 covers day to day

Agent 365 gathers in a single console the inventory of agents, their identity, their activity and the rules that apply to them. Every agent becomes an identifiable object in Microsoft Entra Agent ID, which lets you assign an owner, follow its access and retire it cleanly the day it no longer has a purpose.

Observability sits at the centre: you know what an agent did, which tools it used and which data it read. Security and governance policies then apply at fleet level, without reconfiguring each agent one by one. Audit logs close the loop and provide the evidence expected during an internal review.

The Swiss context: traceability and data protection

In French speaking Switzerland, teams deploying agents often handle data covered by the revised Federal Act on Data Protection, by banking requirements or by medical confidentiality rules. Showing who approved an agent, what it accessed and when it was retired is not administrative comfort, it is a condition for passing an audit.

The course runs in Geneva and Lausanne, on site or in a virtual classroom, with a trainer who answers questions about your own environment. That is the difference with self study: official documentation describes features, it does not tell you which policy to apply to an HR agent reading employee records.

Where it fits among ITTA courses

AB-6007 targets technical profiles who must govern what already exists. If your priority is the business use of generative AI, Transform business workflows with generative AI (AB-730) is a better entry point. For teams who mainly want to get value from Copilot every day, Craft effective prompts in Microsoft 365 Copilot (MS-4005) builds the right writing habits. The three courses complement each other: usage, prompting, then governance.

After the half day

The most useful next step is to build the real inventory of your agents within the following week, then assign a named owner to each one. You define a first minimal policy, switch on log review and set a control frequency. That baseline is usually enough to regain control of a fleet that grew faster than the procedures around it.

FAQ

Who is this half day designed for?

Microsoft 365 administrators, security officers and architects who must supervise agents already used inside the company. The content assumes administration experience, not development expertise.

Do I need agents already deployed to attend?

No, although it helps. The scenarios make far more sense when you have a few agents from your own organisation in mind, even at pilot stage.

How does governing an agent differ from governing a user?

An agent acts fast, often without direct supervision, and chains actions across several services. It therefore needs its own identity, a human owner, a defined lifespan and continuous monitoring of the tools it can reach.

Is half a day enough for such a topic?

The format targets implementation: understanding the Agent 365 model, knowing where to look and applying the first policies. Organisations going further continue with AB-730 or MS-4005 depending on the profiles to train.