Microsoft Purview adds prompt logs for Copilot Studio agents

Purview reached Copilot Studio agents. If your company must keep AI chat records, this is the switch. If you enable it without telling staff, you will buy audit comfort with trust debt.

How it works at a high level

When the tenant enables the connector, prompts and responses from Copilot Studio agents can flow into Purview retention. Policies you already use for other workloads apply. Nothing about the announcement says every tenant wakes up logged by default. Enablement is a choice with paperwork behind it.

Decide the purpose in writing: legal hold, regulated industry rules, incident review. “Because we can” is not a purpose.

People side

Announce the change. Update acceptable use. Remind staff not to paste secrets into agents regardless of logging. Logs make secrets a longer-lived problem.

Compliance tooling should be visible. Microsoft wired the pipe. Your culture decides whether the pipe feels like safety or surveillance. Prefer safety: clear notice, narrow purpose, finite retention.

Retention recipe

Start with short retention—thirty or sixty days—unless counsel demands more. Separate legal-hold projects from ordinary agent logs. Provide a redaction path for prompts that accidentally include customers’ government IDs. Announce a contact for “I pasted something I should not have.” Purview can store. Your program must also forget and fix. Enablement without those paths trains people to bypass the official agents entirely.

Agent design implications

Once logging is on, design agents to ask for the minimum necessary context. Discourage free-text pasting of spreadsheets when a record ID would do. Add soft warnings in the agent greeting: “Do not paste passwords or ID numbers.” Purview makes prompts durable. Durable prompts should be small. Architecture and copy both change when retention is real. Schedule a quarterly access review of who can read those logs, the same way you review mailbox eDiscovery roles.