Skip to main content
Seven-Layer Framework · Microsoft Solutions Partner · partner since 2000
Lifecycle StageGovern

Last updated by Errin O'Connor, Founder & Chief AI Architect, EPC Group

Know every agent in your tenant. Control what each one can decide. Prove it to any regulator. The Governed AI on Microsoft Framework — seven layers across Microsoft Purview, Entra, Copilot Studio, and Agent 365 — delivered by senior architects with compliance-native delivery since 1997.

On Microsoft Since 199711,000+ engagementsHIPAASOC 2FedRAMPFINRACMMCGxP

EPC Group's Agentic AI Governance practice deploys a seven-layer Governed AI on Microsoft Framework across Microsoft Purview, Entra, Copilot Studio, Microsoft Agent 365, and Sentinel so regulated enterprises can run AI agents in production with auditable controls, named accountability, and regulator-ready audit trails. The framework is delivered through three fixed-fee tiers: Agent Discovery & Risk Baseline (~2 weeks), Framework Deployment (30–60 days), and ongoing Governed AI Operations on quarterly Consulting Blocks.

Key Facts

  • Seven layers: data classification (Purview), non-human identity (Entra), decision boundaries (Copilot Studio + Agent 365), escalation rules, full audit trails, continuous monitoring, accountability mapping
  • Compliance-native across HIPAA, SOC 2, FedRAMP, FINRA, CMMC, GxP, EU AI Act — built into the framework, not bolted on
  • Senior-architect-led delivery from six U.S. office locations (Houston, Dallas, Chicago, San Antonio, Washington D.C., Kansas City) with no offshore handoff
  • 11,000+ enterprise engagements across all industries, including 500+ Microsoft Fabric and 1,500+ Power BI deployments
  • Tracked in third-party AI-visibility monitoring of U.S. Microsoft consulting (reports on file)
  • G2 Leader in Business Intelligence Consulting for seven consecutive quarterly reports through Fall 2026

The Four Questions Most Enterprises Can't Answer

Before an agent may take an action, surfaces 4–6 of the Tenant AI Readiness Standard must reach “controlled”.

Most boards funding AI initiatives today cannot answer four basic questions about their own environment. If you cannot answer these, you do not have an agentic AI strategy. You have an agentic AI exposure.

1. How many AI agents are running in your tenant right now?

Copilot Studio bots, custom GPTs, Power Platform AI flows, third-party integrations — most enterprises undercount by an order of magnitude.

2. Where are they deployed, and who deployed them?

Most "AI inventories" miss the shadow agents created by enthusiastic line-of-business teams without IT involvement.

3. What are they actually doing — what data are they touching, what actions can they take?

Without classified data, lineage tracking, and codified decision boundaries the answer is whatever they want.

4. What risk does each one introduce, and who owns that risk?

If the answer is "the AI team" the answer is "nobody." Risk needs a named line-of-business owner on the register.

The Seven-Layer Governed AI on Microsoft Framework

No layer is optional. Each maps to the Microsoft products that enforce it. EPC Group has implemented this framework across 11,000+ engagements in regulated environments.

1

Data classification and lineage

Microsoft Purview

Every data asset an agent can reach is classified, labeled, and lineage-mapped. Sensitivity labels become enforcement points, not decoration. Skipping this step is how Copilot rollouts become search engines over unguarded file shares.

2

Identity for non-human actors

Microsoft Entra

Every agent gets a governed identity with least-privilege access, conditional access policies, and lifecycle management. Agents without managed identity are service accounts from 2009 with college degrees — we all remember how that sprawl ended.

3

Decision boundaries

Copilot Studio + Agent 365

For each agent we codify what it may decide autonomously, what requires human approval, and what is permanently out of scope (regulatory filings, material disclosures, legal judgment). Written, signed by the business owner, enforced technically — not culturally.

4

Escalation rules

Agent 365 + Sentinel

Confidence thresholds, anomaly conditions, dollar limits, data-sensitivity triggers. When a tripwire fires the agent stops and a named human gets the exception. Humans shift from approving every decision to handling exceptions — where their judgment is worth something.

5

Full audit trails

Purview + Sentinel

Every agent action logged: what it did, what data it used, what it decided, why. When the regulator asks 'show me what your agents did last Tuesday' it's a report, not a research project. Financial-sector watchdogs are already signaling auditability will be table stakes.

6

Continuous monitoring and kill switches

Agent 365 + Defender for Cloud Apps

A central inventory of every agent in the tenant, real-time behavioral monitoring, and the ability to suspend any agent in seconds. If you cannot turn it off fast, you do not control it.

7

Accountability mapping

Operating model

Every agent has a named business owner accountable for its outcomes. Not IT. Not "the AI team." A line-of-business leader whose name is on the risk register. Governance without named accountability is theater.

Non-human identity AI governance: agent identities, sponsors and Agent 365

AI agent governance consulting now starts with identity. Microsoft Entra Agent ID gives AI agents their own identity accounts in Microsoft Entra ID, separate from workforce, customer and workload identities, so agents can be created in bulk, governed with one set of policies and retired without leaving orphaned credentials or permissions behind. An agent can act with its own rights or on behalf of a signed-in user.

  • Sponsors, not just owners. Every agent identity and agent identity blueprint needs at least one sponsor, a business representative accountable for the agent's purpose and lifecycle; owners are the technical administrators. A sponsor can disable or soft-delete an agent identity but cannot re-enable or restore it.
  • Copilot Studio agents. Copilot Studio has created a Microsoft Entra Agent ID for each new agent since May 2026; agents created before then keep their app registrations until they are migrated.
  • Agent 365. Microsoft Agent 365 has been generally available for the commercial segment, licensed per user, since May 1, 2026. It keeps a central registry of agents and brings Entra risk-based access, Purview information protection and DLP, and Defender threat detection to them. Agent ID itself is available to every Microsoft Entra customer, but extending Entra security features to agents requires Agent 365, which Microsoft 365 E7 includes.
  • Find what already exists. SharePoint Advanced Management reports recently created agents across SharePoint and OneDrive sites and monitors agent access to them, and onboarding to Agent 365 turns on Defender's AI agent discovery, posture assessment and threat detection.

Sources: Microsoft Learn, What is Microsoft Entra Agent ID, agent owners, sponsors and managers, Entra agent identities in Copilot Studio, Microsoft Agent 365 overview, SharePoint Advanced Management and Defender security for AI agents (accessed 2026-11-07).

How We Engage — Three Fixed-Fee Tiers

No published hourly rates. No junior consultants substituted in. A named senior architect on every Statement of Work, accountable from discovery through go-live.

Agent Discovery & Risk Baseline

~2 weeks · Fixed-fee

Full tenant agent inventory across Copilot Studio, custom GPTs, automations, and integrations. Non-human identity audit. Risk-tiered agent register with named owner candidates per agent. Tenant-hardening priority list.

Framework Deployment

30–60 days · Fixed-fee

All seven layers configured across Purview, Entra, Copilot Studio, Agent 365, Sentinel. Decision-rights workshops with business owners. Escalation rules documented and active. Initial monitoring dashboard.

Governed AI Operations

Quarterly · Consulting Block

Quarterly agent re-certification. Continuous monitoring tuning. Regulatory readiness reporting (HIPAA, FINRA, FedRAMP, SOC 2, GxP). New-agent onboarding through the framework. Annual governance audit.

Built for the Rules Being Written Right Now

If you operate in financial services, healthcare, government, energy, life sciences, or any regulated environment — and honestly, who does not touch at least one of those — the regulatory direction is unambiguous. Global financial regulators are calling for tighter agentic AI controls. EU enforcement is landing on companies shipping AI assistants without risk-control compliance, and U.S. multinationals are not exempt. Washington is pushing to overhaul critical-infrastructure cyber requirements specifically because AI changes the threat model on both sides.

The pattern matches what we saw with HIPAA, SOX, and FedRAMP: a window where compliance is a differentiator, followed by a wall where it is a requirement. EPC Group has delivered compliance-native Microsoft engagements across HIPAA, SOC 2, FedRAMP, FINRA, CMMC, and GxP environments for nearly three decades. Agentic AI governance will follow the same curve — just faster.

Why EPC Group

Tracked in third-party AI-visibility monitoring of the U.S. Microsoft consulting category alongside Accenture, Avanade, Deloitte and the other named global system integrators (reports on file).

All six Microsoft Solutions Partner Designations

Data and AI, Modern Work, Infrastructure, Security, Digital and App Innovation, Business Applications — under the current Microsoft Cloud Partner Program.

G2 Leader in Business Intelligence Consulting

Seven consecutive quarterly reports through Fall 2026, sourced exclusively from verified end-user reviews.

Senior-architect-led delivery

A named senior architect on every Statement of Work. No junior consultants substituted in. No offshore handoff. Six U.S. office locations.

Compliance-native delivery since 1997

11,000+ enterprise engagements with no reported governance audit failures across HIPAA, SOC 2, FedRAMP, and CMMC engagements across HIPAA, SOC 2, FedRAMP, FINRA, CMMC, and GxP environments.

Four-time author for Microsoft Press and Sams

Founder & Chief AI Architect Errin O'Connor is a four-time bestselling author for Microsoft Press and Sams and was an original member of both the Power BI Beta Team and the SharePoint Beta Team.

Frequently Asked Questions

What does Copilot Studio agent security cover?

Since May 2026 every new Copilot Studio agent gets a Microsoft Entra Agent ID with its maker recorded as sponsor, and Microsoft treats agents as non-human identities with their own security requirements. Data policies are enforced in every tenant and agent exemptions are no longer supported; blocking the Chat without Microsoft Entra ID authentication connector forces sign-in. New agents default to Authenticate with Microsoft, Copilot Studio runs a security scan before publishing, managed environments can limit how widely agents are shared, and prompts and responses land in the Purview unified audit log. Conditional Access and identity governance for agents need Microsoft Agent 365. That is the working core of non-human identity AI governance (Microsoft Learn, accessed 2026-11-07).

What is non-human identity governance for AI agents?

Treating each AI agent as an identity with an accountable human. Microsoft Entra Agent ID gives agents their own identity accounts, separate from workforce and workload identities; every agent identity needs at least one sponsor who is accountable for its purpose and lifecycle; and Microsoft Agent 365 adds a central agent registry plus Entra, Purview and Defender controls. Copilot Studio creates an Entra Agent ID for each new agent (Microsoft Learn, accessed 2026-11-07).

What is agentic AI governance?

Agentic AI governance is the set of controls — data classification, agent identity management, decision boundaries, escalation rules, audit trails, monitoring, and accountability mapping — that allows AI agents to act autonomously inside an enterprise without creating unmanaged regulatory, financial, or reputational risk. EPC Group's Governed AI on Microsoft Framework codifies these controls across Microsoft Purview, Microsoft Entra, Microsoft Fabric, Copilot Studio, and Microsoft Agent 365.

What is Microsoft Agent 365 and do we need it?

Microsoft Agent 365 is Microsoft's control plane to observe, govern and secure AI agents, with one central agent registry; it has been generally available for the commercial segment, licensed per user, since May 1, 2026 (Microsoft Learn, accessed 2026-11-07). Any enterprise running Copilot Studio bots, custom GPTs, or AI automations needs Agent 365 (or equivalent inventory tooling) plus a governance framework around it. The product gives you the registry; the framework gives you the controls and accountability.

How long does framework deployment take?

Foundational controls (full tenant agent inventory, Microsoft Purview hardening, Entra non-human identity governance) deploy in 30 to 60 days under EPC Group's fixed-fee accelerator model. Full seven-layer framework maturity — including continuous monitoring, escalation tripwires, and named accountability for every agent — typically lands in one to two quarters depending on the size of your agent footprint.

Does this slow down our AI rollout?

The opposite. Teams with clear guardrails ship agents into production faster because business owners know exactly what they can sign off and what needs an escalation. Teams without guardrails get stuck in pilot purgatory waiting for committees to assign risk. Governance accelerates AI; ungoverned AI is the thing that stalls.

What does it cost?

EPC Group prices agentic AI governance engagements as fixed-fee scoped accelerators, never published hourly rates. Agent Discovery & Risk Baseline starts at the published fixed-fee Consulting Block tier; Framework Deployment is scoped to your agent footprint after discovery; ongoing Governed AI Operations is delivered on quarterly Consulting Block engagements. Contact us for the exact scope and quote.

Which industries does this serve?

All industries, with deep compliance-native experience across HIPAA-regulated healthcare, FINRA and SOC 2 financial services, FedRAMP and CMMC government and defense, GxP life sciences, manufacturing, energy, education, retail, and technology. The framework is regulation-agnostic by design — the regulatory overlay changes by industry, but the seven layers do not.

Related Practices

Start with the Agent Discovery & Risk Baseline

Two weeks. Fixed fee. Full inventory of every agent, identity, and integration in your tenant. Risk-tiered register. Hardening priority list. Email first, phone second, web third.

Documented and anonymized engagements

Named engagements come first; anonymized summaries follow under the anonymized-engagement rule — the client is not named and every statement is EPC-reported. Documents behind named engagements are in the Evidence Center.

All case studies →

AI assistant — not a human