Skip to content
TheGovernor.

Pillar 01 / 07

AI and agent control

Which agents exist, who can talk to them, what can they reach and whose credentials do they use?

Who asks
CIO, CISO, AI lead
The outcome
Copilot and agents that only see what they should.

“Before you roll out Copilot and agents, fix who can see what. Otherwise Copilot surfaces every overshared file, and agents get built that nobody controls.”

Why now

Every Microsoft customer is under pressure to adopt AI, and the EU AI Act now asks who answers for each agent.

What it covers

Copilot Studio agents and the settings around them: authentication, sharing, channels, tools and the credentials they run on, MCP servers, knowledge sources and generative AI tenant settings.

What The Governor checks

Measured on every scan and scored for this pillar alone.

  • Agents in the inventory, and the share living in the default environment
  • Published agents anyone can talk to, without sign-in
  • Agents shared with the whole tenant
  • Agent tools that run on the maker's own credentials
  • MCP tools connected to agents
  • Agents with open knowledge sources
  • Copilot data sharing and logging settings

What our team adds

The parts no scan can settle: decisions, design and rollout.

  • An AI readiness plan for Copilot and agents, built on your own scan
  • Oversharing clean-up before Copilot reads your files
  • Agent data policies, publishing rules and an intake gate

What a first scan often shows

Examples from a demo tenant.

  • Critical3 published agents answer without authentication
  • High5 agent tools send email on the maker's account
  • Medium11 agents live in the default environment

Find out where you stand, pillar by pillar.

Start the trial on your own tenant, or let a solution architect read the first scan with you.