Monitoring Copilot with Communication Compliance

I’ve found Communication Compliance is one of the least understood aspects of Purview, and as a consequence it’s one of the least implemented. That’s a shame because for not much configuration work, you get a significant addition to your security and compliance posture. Even when it is discussed, its value for Copilot tends to be overlooked compared to labels or DLP. Microsoft’s own Zero Trust workshop guidance calls this out as a blind spot if not enabled. If you monitor email and Teams but not Copilot, you have an unmonitored channel that bypasses your compliance controls.

The good news is that closing the gap doesn’t take long. Communication Compliance can analyze the prompts people send to Copilot and the responses they get back, and a built-in template does most of the work.

Licensing Note: Communication Compliance is licensed per user through plans such as Microsoft 365 E5, Office 365 E5, Microsoft Purview Suite, or the E5 Insider Risk Management add-on. Analyzing Microsoft 365 Copilot interactions carries no pay-as-you-go charge, but Microsoft pairs that capability with a Copilot license. Monitoring other AI systems requires pay-as-you-go billing.

The Value of Communication Compliance

Communication Compliance has long been the tool for catching harassment, threats, regulatory violations, and leaks of confidential information in email, Teams, and Viva Engage. Pointing it at Copilot extends the same standard to AI-generated content.

The Copilot template adds two detections you won’t get from a standard policy. Prompt Shields looks for prompt injection attempts, often called jailbreaks. The protected material classifier flags responses that contain branded or copyrighted content. You can layer in the usual classifiers for threats, discrimination, and sensitive information. In my experience, the protected material classifier isn’t as useful as prompt injection, but I still suggest enabling both for clients.

Privacy is built in. Reviewers with the Analyst role see pseudonymized usernames, and only Investigators see real names. Investigators have to be opted in by an administrator.

Setting It Up

Microsoft’s Copilot configuration guide walks through the steps, so I won’t repeat them. In short, go to Communication Compliance > Policies > Create policy and choose the Detect Microsoft 365 Copilot and Microsoft 365 Copilot Chat interactions template. Pick your users and reviewers, then create the policy. A few things to watch for:

• Roles come first. You need to be in one of the Communication Compliance role groups before the menu even appears, and permissions can take up to 30 minutes to apply.
• Reviewers must be individuals. They need Exchange Online mailboxes and the Analyst or Investigator role for Communication Compliance, and groups aren’t supported as reviewers. This is one Purview policy where HR or Legal should be involved, not just IT.
• Scoped-user groups have limits. Dynamic distribution groups, nested groups, Microsoft 365 Groups with dynamic membership, and shared mailboxes aren’t supported for scoped users.
• Not retroactive. The new policy gathers signals every hour starting from when you create the policy.
• Auditing must be on. It’s the default, but confirm it hasn’t been disabled.
• No PowerShell. Policies can’t be created or managed that way, so build them in the portal.
• Start broad. Review all interactions first to learn how people actually use Copilot, then tune to reduce noise. Leave the review percentage at 100%, and select all the trainable classifiers to start.
• SIEM integration. If you run a SIEM, the policy alerts can be sent there. At a minimum, they are in Purview’s audit logs.

Reviewing Matches

Flagged interactions show up on the policy’s Pending tab, with prompts and responses listed as separate entries. If both match, you’ll see two items, and if only one matches, you’ll see just that one. For Microsoft Copilot interactions, a Copilot icon and a [Copilot] value with the app name in the Subject column identify them. Selecting an item shows the full message text that caused the match. You remediate these matches the same way as any other Communication Compliance policy match, and they’re included in Communication Compliance reports and audit data. This is one of the places in Purview where you can see Copilot prompt and response text, though only for content that matches a policy.

Integration with Insider Risk Management

Communication Compliance and Insider Risk Management are designed to work together. Insider Risk Management data can enrich the triage experience, so investigators see a user’s risky activities and risk severity alongside the alert. In the other direction, Communication Compliance policy indicators can feed Insider Risk Management policies, including jailbreak alerts.

The Risky AI usage policy template detects sensitive information in prompts and responses, plus browsing to generative AI sites, and feeds Adaptive Protection risk levels. The template can use an HR connector and Communication Compliance risk signals, but neither is required, so a Communication Compliance policy is optional, though it widens coverage. I think Insider Risk without Communication Compliance is an incomplete solution.

Where to Start

If you haven’t yet created a Communication Compliance policy for Copilot (and other Microsoft 365 workloads) and have the licenses, it’s a quick win with significant reporting benefits. If you need to, start with a pilot group, but don’t let planning drag on. I hope this helps someone get their first policy running.