Communication Compliance Copilot lets you review Copilot prompts

A question I keep getting at the moment is how an organisation can introduce Microsoft 365 Copilot without losing track of sensitive or inappropriate content. In larger companies the works council will want to know exactly what is monitored, who does the monitoring and which privacy safeguards apply. Communication Compliance Copilot in the Purview portal means Microsoft now covers generative AI prompts and responses as well as the classic communication channels. If you were hoping Copilot could run quietly under the radar, that won't happen. This has become a governance topic you can't skip.

Communication Compliance Copilot reviews prompts and responses at the source

Data Security Posture Management (DSPM) for AI looks at data and model usage. Communication Compliance Copilot works one step earlier and reviews the text users send to Copilot along with the responses they get back. In practice a reviewer team can be alerted to individual Copilot interactions when a policy detects “inappropriate language”, “disclosure of confidential info” or some other trigger. If you have built policies for Teams or email before, the review pipeline will feel familiar. The difference is that here the content is AI-generated or shaped by AI.

In many tenants I work with, a close partnership between security and the works council makes all the difference. Agree on what you'll monitor before you start instead of arguing about it afterwards.

You create the policies in the Purview portal under Communication Compliance. That's where the generative AI templates live, and it's also where you assign reviewers, set the scope (pilot groups, for example) and, most importantly, configure the privacy options. I wouldn't run Copilot with real guardrails without this capability.

The monitoring flow runs from the user's prompt to the reviewer team

The diagram below shows the process in simplified form:

flowchart TD
User["User sends Copilot prompt"]
Copilot["Copilot processes prompt"]
Result["Copilot returns response"]
Compliance["Communication Compliance reviews prompt & response"]
Reviewer["Reviewer team notified"]
User --> Copilot --> Result --> Compliance --> Reviewer

Figure: How Communication Compliance reviews a Copilot interaction (simplified).

The policies rely on the same core detection as the other channels, but you can adjust them for generative AI (Microsoft's documentation).

The setup combines policies, reviewers, a pilot scope and privacy controls

You start in the Purview portal under Communication Compliance → Policies. Since early 2026 Microsoft has shipped a dedicated template for Copilot prompts and responses that detects toxic language, sexual content and compliance risks. Next you assign reviewers, for example your compliance team or a mix of people from security and HR. My strongest advice is to begin with a small pilot scope. Rolling it out to the whole company on day one is rarely a good idea.

The privacy controls matter a great deal. Usernames can be pseudonymised for cases sent to review, and you can limit which reviewers see raw content. Being completely open with the works council is not optional. I cover the technical and organisational side of that conversation in Copilot works council: Technical controls in practice.

This is what a typical policy looks like (created in the Purview portal under Communication Compliance → New policy):

• Template: Generative AI
• Channel: Copilot prompts and responses
• Scope: Pilot group (e.g. IT and HR)
• Reviewers: Compliance team
• Privacy: Pseudonymisation enabled

Matches show up in the review dashboard as usual. Reviewers can compare them against training data there and take whatever action fits.

Communication Compliance Copilot does not replace DSPM for AI

Communication Compliance Copilot won't trace AI model usage, prompt injection or shadow AI threats. Other Purview features (DSPM for AI) handle those scenarios, with a focus on data classification, permissions and oversight at the level of the model and the data flow. If you need to know which sensitive data Copilot could leak to external destinations, Communication Compliance Copilot on its own isn't enough. That job belongs to DLP and DSPM rules.

One mistake I see regularly is expecting Communication Compliance Copilot to cover every AI risk. What it does well is review content against pre-defined (or custom) patterns. It doesn't govern all data movement and it doesn't block anything in real time. Keep expectations realistic from the start and you'll save yourself some misunderstandings.

I treat Communication Compliance Copilot as one building block of governance

If your organisation plans to roll out Copilot with genuine governance and privacy requirements, Communication Compliance Copilot is a useful piece of the puzzle. It is not a replacement for a complete DLP or DSPM strategy. In my projects a phased approach has worked best. Start with pilot groups, go through the policies with the works council at regular intervals and keep the reviewer pool small and well trained.

Most of all, set goals you can actually meet. Communication Compliance Copilot detects risky AI text, raises alerts and helps you manage it, but interpreting the findings and acting on them is still the organisation's job. Use it as a governance tool rather than a surveillance machine and it will help you get productive, compliant use out of Copilot.