Defender + Entra + Purview agent governance for HIPAA, FINRA, SEC, FedRAMP, CMMC, GxP, and EU AI Act — field-tested implementation playbook by EPC Group
By Errin O'Connor, Founder & Chief AI Architect, EPC Group
Microsoft Agent 365 (generally available May 1, 2026 either inside Microsoft 365 E7 at $99/user/month or as a $15 standalone add-on) is the governance control plane for AI agents across the Microsoft enterprise tenant. For regulated industries — healthcare HIPAA, financial services FINRA + SEC, federal FedRAMP + CMMC, life sciences GxP, and EU AI Act-impacted operations — Agent 365 configuration must happen BEFORE broad Microsoft Copilot Studio, Azure AI Foundry, or third-party agent deployment. This playbook walks through EPC Group's field-tested Agent 365 governance configuration across the four Agent 365 admin surfaces: Microsoft Defender (posture + threat protection), Microsoft Entra (agent identity + access), Microsoft Purview (agent data governance), and the Microsoft 365 admin center (agent inventory + lifecycle).
For 25 years, the Microsoft enterprise control plane has relied on three main pillars:
Every audit narrative — HIPAA, FINRA, SEC, FedRAMP, CMMC, GxP — is based on these pillars and connects to one key element: the named human user.
Microsoft Copilot, Microsoft Copilot Studio, Microsoft Security Copilot, and the latest third-party agents have transformed the traditional model. These include Claude Code, GitHub Copilot CLI, and OpenClaw.
Unlike humans, these agents do not authenticate as named identities. They can consume credentials at scale and access enterprise data much faster than a person can.
Additionally, they generate communications such as:
These communications previously existed in a regulatory gray zone before Agent 365.
Agent 365 will be available to everyone starting on May 1, 2026. You can find it in Microsoft 365 E7 for $99 per user per month. Alternatively, it can be purchased as a standalone add-on for $15.
For regulated industries, implementing Agent 365 is not just about configuration. It is also about meeting regulatory requirements.
Agent 365 aligns with the four Microsoft admin responsibilities that already structure the enterprise control plane.
EPC Group's regulatory-baseline matrix shows what Agent 365 must enforce per industry. Every baseline below ties to a specific statutory control reference.
PHI sensitivity labels propagate to agent grounding; Audit Premium 7-year retention; agents bound to BAA-verified tenant scope; Microsoft Restricted SharePoint Search for agent grounding scoped to allowlisted clinical and administrative sites
MNPI sensitivity labels enforced at agent grounding; FINRA Rule 4511 prompt logging in AI Hub; SEC Rule 17a-4(f) tamper-evident retention; Communication Compliance for FINRA 3110 supervisory review on agent-generated communications; Information Barriers separate research and investment banking agent populations
Microsoft 365 GCC High deployment; CUI-aware sensitivity labels; NIST 800-53 + 800-171 control mapping; CMMC 2.0 Level 2 / Level 3 alignment with IL4 / IL5 boundary enforcement; agent identities scoped to authorization boundary
21 CFR Part 11 and FDA Annex 11 validated agent workloads; clinical-trial data isolation; Audit Premium retention tied to the regulatory clock for the relevant clinical phase; change-control records for every Agent 365 policy revision
Annex III high-risk classification assessment for agents touching education, employment, law enforcement, or migration data; GDPR Article 32 technical and organizational measures via Purview + Entra; data-residency enforcement on agent grounding paths
Customers assign E7 licenses or the $15 add-on, but never actually configure the agent inventory, posture policies, sensitivity-label propagation, or Conditional Access for agents. The license is purchased; the governance is theoretical.
Without the regulatory-baseline confirmation step, agents end up grounded against PHI, MNPI, or CUI without enforcement. The first regulator audit then becomes a remediation project on top of an in-flight rollout.
Customers stand up the sanctioned agents but never deploy Defender + Intune shadow-agent discovery. Three months later they find 200 unmanaged Claude Code installations across endpoints — and have no policy to enforce.
Agent 365 governance gets configured once and never reviewed. Within six months drift erodes the posture. The fix: quarterly Compliance Manager attestation with the customer's named Chief Compliance Officer or equivalent.
Pre-Agent 365, AI agents authenticated as service principals or app registrations — a model designed for back-office integrations, not for end-user-facing AI assistants. Auditors and regulators have been increasingly uncomfortable with this gap because (a) you cannot meaningfully attribute a specific agent action to a specific identity for incident response, (b) Conditional Access cannot reason about agent intent, and (c) Communication Compliance and Insider Risk Management have no agent-specific signal. Agent 365 closes the gap by giving every agent a named identity, an inventory record, a posture profile, and a data-governance contract. For HIPAA, FINRA, SEC, FedRAMP, CMMC, and GxP environments — where every agent interaction must be attributable, retained, and auditable — Agent 365 is foundational.
EPC Group's Governed AI on Microsoft Framework is a seven-layer methodology (Identity / Data Classification / Data Plane / Model Governance / Prompt + Output Controls / Audit Retention / Continuous Improvement). Agent 365 augments Layer 1 (with agent identity in Entra), Layer 3 (with agent-aware DLP signals in Purview), Layer 5 (with the Purview AI Hub agent prompt + response capture), Layer 6 (with Audit Premium retention covering agent activity), and Layer 7 (with Defender agent posture + threat hunting). The framework as a whole stays intact — Agent 365 is the agentic-AI augmentation we have been waiting for since the first Copilot Studio deployments.
EPC Group ships Agent 365 in three phases. Phase 1 Readiness Assessment (4-6 weeks, fixed-fee): Defender + Entra + Purview + admin-center current-state audit, regulatory baseline confirmation, agent inventory (existing + planned), and a phase-2 scoping recommendation. Phase 2 Foundation (10-14 weeks, fixed-fee): agent identity model in Entra, Conditional Access policies for agents, sensitivity-label propagation to AI Hub, Audit Premium retention configured, agent inventory and lifecycle workflows in Microsoft 365 admin center, Defender agent posture baseline. Phase 3 Enterprise Scale (16-26 weeks, fixed-fee): Information Barriers for agents, Communication Compliance for agent-generated communications, Sentinel detection rules for agent anomalies, multi-tenant federation where applicable (GCC + commercial + GCC High), quarterly Compliance Manager attestation rhythm.
Microsoft Defender + Microsoft Intune introduce new capabilities for discovering shadow agents — AI tools installed on Windows endpoints outside the sanctioned Agent 365 inventory. The first wave covers OpenClaw with expansion to GitHub Copilot CLI, Claude Code, and other commonly-installed local agents. Discovery is paired with policy controls: block unsanctioned agents, redirect to sanctioned alternatives, or place in a quarantine zone with elevated monitoring. For regulated industries this is critical — every shadow agent is a potential data-exfiltration vector and a potential compliance finding. EPC Group integrates shadow-agent discovery into Phase 2 of the Agent 365 engagement.
For regulated tenants the right answer almost always factors in the Entra Suite inclusion. If you already use Microsoft Entra Identity Governance access reviews, Verified ID, Private Access, or Internet Access, E7 captures all four products at a marginal cost. If your tenant is already on E5 with broad Copilot adoption planned and regulator-grade agent governance required, E7 beats E5 + standalone Agent 365 + standalone Entra Suite. If you are a smaller regulated tenant with constrained Copilot adoption and existing Entra Identity Governance posture, standalone Agent 365 at $15 may be sufficient. EPC Group's Readiness Assessment includes per-tenant E7-vs-standalone economic modeling.
Fixed-fee Readiness Assessment (4-6 weeks). Tenant-specific regulatory baseline + Agent 365 configuration roadmap. Senior architects (not sales) take discovery calls.
AI assistant — not human