Power Platform Engineer cover image. Category: Copilot Studio. Title: Data policies bind every agent. Check what they name.
,

Copilot Studio Data Policies Now Apply to Every Agent. Check What They Actually Name.

Microsoft Learn updated Configure data policies for agents on 22 October 2026. The page is the reference for how Power Platform data policies, the controls many teams still call DLP, apply to Copilot Studio agents. Its governance point is blunt: enforcement covers every agent in every tenant, exemptions are gone, and a tenant-scoped policy reaches environments that do not exist yet. The same page also shows the limit of that control. It names a specific set of Copilot Studio connectors, and several capabilities makers now use every day do not appear on it.

Power Platform Engineer read the page in full today, along with its troubleshooting companion and the connector endpoint filtering article it relies on. This piece sets out what a data policy can block, how enforcement shows up for makers, users and admins, and where the documentation stops.

The capability at a glance

AreaWhat the Learn page states
Enforcement“Since early 2025, data policy enforcement is in effect for all tenants,” citing message center post MC973179.
Exemptions“Agent data policy enforcement exemption is no longer supported.” Previously exempted agents are now subject to enforcement.
TimingCopilot Studio “supports data policy enforcement in real time.” Makers and users see error messages for violations.
Who can configureA tenant admin or a user with the Environment Admin role.
WherePower Platform admin center, Security, Data and privacy, Data policy.
Scope“Policies with a tenant scope apply to all agents in all environments across the tenant.” Add all environments also applies to new environments.
Licensing or statusThe page carries no preview or GA label and does not mention licensing or Managed Environments.

What a data policy can actually block

The page lists the connector names that appear in the Power Platform admin center for Copilot Studio. Each one maps to a maker capability, and classifying it as Blocked, or placing it in a different data group from the agent’s other connectors, is how an admin switches that capability off.

  • Chat without Microsoft Entra ID authentication in Copilot Studio: blocks publishing agents that are not configured for authentication. Once a policy requires it, No authentication and some manual authentication options are not available to makers.
  • Prebuilt and custom connectors: blocks Power Platform connectors used as tools. The page adds that this “also blocks access to tools in connected MCP servers, which rely on Power Platform connectors for connectivity.”
  • HTTP: blocks publishing agents that make HTTP requests, with endpoint filtering for finer control.
  • Channel connectors: Direct Line channels, Microsoft Teams + Microsoft 365, SharePoint, Facebook, WhatsApp and Omnichannel, each listed separately.
  • Microsoft Copilot Studio: blocks event triggers and automated evaluations run with authenticated accounts.
  • Skills with Copilot Studio: blocks Azure Bot Service skills.
  • Application Insights in Copilot Studio: blocks connecting agents to Application Insights.
Diagram of the eight things a Copilot Studio data policy can block or enforce: user authentication, connectors as tools including MCP server tools, HTTP requests, channels, event triggers, Azure Bot Service skills, Application Insights, and no exemptions since March 2025.
The Copilot Studio connectors a data policy can block, as named on Microsoft Learn.

One detail deserves attention before any policy change. The page warns that connectors introduced after 2019, including Chat without Microsoft Entra ID authentication and Direct Line channels, are “likely to be part of the default ‘Non-business’ group,” and that “in many organizations, connectors in the ‘Non-business’ group are automatically blocked.” The troubleshooting article is more direct: data policies that block new connectors by default “could result in production outages,” and Direct Line or unauthenticated website deployments “might be unexpectedly blocked.” The default data group is therefore a production setting, not a housekeeping one.

The skills naming trap

The page opens with an Important note that the skills it governs are “Azure Bot Service skills, a code-based way to extend the capabilities of an agent in the standard harness,” which “shouldn’t be confused with agentic skills in the Microsoft Copilot harness.” That matters because the Skills with Copilot Studio connector sounds like the obvious way to stop makers adding skills. On the evidence of this page, it is not. Organizations that have started sharing agentic skills, covered in Copilot Studio Skills Can Now Be Shared Across Your Organization, need a separate review path for them.

How enforcement shows up

Enforcement is visible in three places, and each audience needs to know what it looks like.

  • Makers see blocked connectors disabled in the tool picker and restricted channels disabled when publishing. “When a data policy violation occurs, the Publish button becomes unavailable.” A skill that is blocked returns an error suggesting the maker contact an admin, and blocked event triggers show an error in the Triggers section of the Overview page.
  • Users of a published agent in violation see an error ending in “Error code: DataLossPreventionViolation” and are told to contact their admin. The agent status reads “You have errors in your published agent.”
  • Admins and makers can open the error details on the Channels page and select Download to retrieve an Excel workbook with two sheets, DLP violations and Blocked channels, including the policy name, policy ID, error type and connector data group.
Diagram showing who sees data policy enforcement: makers see disabled options and an unavailable Publish button, users see error code DataLossPreventionViolation, admins download the Channels workbook, and the Learn page does not name agentic skills, knowledge sources, Work IQ, Copilot connectors or model choice.
Where enforcement appears for makers, users and admins, and what the page leaves out.

The troubleshooting page also confirms that “since March 2025, agents can’t be exempted from data policy enforcement,” and that the PowerShell exemption command no longer works. Any runbook that still relies on an exemption should be retired. If every channel an agent uses is blocked, the agent cannot be published at all.

Governance consequences

First, a tenant-scoped policy is the widest lever an admin has. Because Add all environments “automatically applies to any new environment created in the tenant,” a newly created environment inherits the agent restrictions on day one. That is useful, and it also means a single misclassified connector can break agents in environments the policy owner never looked at.

Second, data policies are about connectors and channels, not identity or model choice. They can force Entra ID authentication for chat, but they say nothing about which identity an agent acts under once it is running. Teams working through the Entra Agent ID migration still need that work, and model restrictions live in separate settings described in Which Model Can a Copilot Studio Agent Use?.

Third, the error message is part of the user experience. The page shows how to set a custom Learn more link and admin contact email for policy errors with PowerShell, and it warns that “these settings apply to all Power Platform apps within the specified tenant.” Agree the wording with the wider platform team before changing it.

What the documentation does not cover yet (as of 15 November 2026)

  • Microsoft Copilot harness capabilities. Apart from the skills note, the page does not mention agentic skills, Work IQ, Copilot connectors as knowledge, or model selection, so it does not say whether data policies govern them.
  • Knowledge sources as a category. The configuration steps mention endpoint filtering for SharePoint or public websites used as knowledge, but the connector endpoint filtering article, last updated 16 June 2026 and labelled “This is a preview feature,” lists only HTTP, HTTP with Microsoft Entra ID, HTTP Webhook, SQL Server, Azure Blob Storage, SMTP, Browser Automation and UI Automation.
  • Licensing, Managed Environments requirements and a preview or GA label. None appear on the page.
  • Recovery after a fix. Neither page says whether a published agent in violation recovers automatically when the policy changes, or whether it must be republished.
  • Tenant-wide inventory. The only discovery method described is the per-agent Channels workbook. No PowerShell, CoE Starter Kit or admin center report for affected agents is named.
  • Cmdlet naming. The text says to use Set-PowerAppDlpErrorSettings, while the sample script uses New-PowerAppDlpErrorSettings. Test both before scripting.

Admin checklist

  • Check the default data group in every tenant and environment policy, and confirm where Chat without Microsoft Entra ID authentication and Direct Line channels currently sit.
  • Decide which channels each environment tier may publish to, and leave at least one allowed channel where agents must be published.
  • Decide whether HTTP is allowed, and if so define endpoint filtering rules, remembering that the feature is in preview.
  • Treat blocking Power Platform connectors as also blocking MCP server tools, and tell makers before you do it.
  • Retire any process that depends on agent exemptions.
  • Before a policy change, ask makers to download the Channels workbook for production agents and review the published agent status afterwards.
  • Set the policy error Learn more link and admin contact email, knowing the setting applies tenant wide.
  • Keep a separate control for agentic skills, knowledge sources and model choice. Do not assume a data policy covers them.

Sources