Copilot Readiness: Fix Oversharing Before You Turn It On

The problem: Copilot makes old mistakes easy to find

Microsoft 365 Copilot grounds its answers in data that the user is already allowed to see. Microsoft's deployment guidance puts it as grounding responses in data users already have permission to access. That is the right design, and it is why oversharing matters so much. Most companies have files that are open too widely: a folder shared with everyone in the company years ago, a site whose owner left, a sharing link that never expired. Nobody found those files because nobody searched for them. Copilot searches for the user, in plain language, in seconds.

Microsoft says the same thing about its own tools. The Purview documentation notes that, because of the power and speed AI can proactively surface content that might be obsolete, over-permissioned or lacking governance controls, generative AI amplifies the problem of oversharing data. So the readiness question is not whether to trust Copilot. It is whether your permissions deserve to be trusted.

Step 1: find where the exposure is

Start with reports, not guesses. SharePoint admin center has Data access governance reports, which are meant to help you discover sites that contain potentially overshared or sensitive content. The reports Microsoft lists are of two kinds.

• Snapshot reports: a site permissions report across the organization, which helps find sites with the broadest access such as sites with thousands of users, external guests or the built-in group Everyone except external users; a site permissions report for a named user; a report of sites and files shared through the special groups Everyone and Everyone except external users; and a report of sites that hold files with specific sensitivity labels.
• Activity reports, which cover the last 28 days: a sharing links report (including Anyone, People in the organization and Specific people links) and a report on sites shared with Everyone except external users.

Microsoft's advice is to start with snapshot reports for a baseline and run them quarterly, then use activity reports monthly to catch new exposure. The documentation says these reports require the prerequisites for SharePoint Advanced Management; it also says that organizations without it must enable data collection for activity reports, that the data is stored for 28 days, and that reports contain only data from when collection was turned on. Licensing terms change, so confirm yours with Microsoft or your licensing adviser.

Then add Purview. The current Data Security Posture Management (DSPM) solution, found in the Microsoft Purview portal under Solutions, includes data risk assessments, under Discover, that help identify and fix potential oversharing. The page says a default assessment runs weekly on the top 100 SharePoint sites based on usage in your organization, with a four-day delay before the first results appear. It summarizes items found, sensitive data detected and the number of links sharing data with anyone. Selecting a site gives you tabs for Overview, Identify, Protect and Monitor.

Custom assessments let you choose users and sources. Optional item-level scanning for Microsoft 365 needs a registered Entra application, applies to SharePoint sites only, and flags items as potentially overshared when they have a sharing link for external or anonymous users, showing any label and the owner. The documented limits are worth planning around: a maximum of 200,000 items per location, no item-level scanning for OneDrive, and a current maximum of 10 SharePoint sites for item-level scanning. A custom assessment's results expire after 30 days, so export what you need. Exports are available as Excel, CSV, JSON and TSV.

A naming note, because Microsoft renamed things. The earlier solution, DSPM for AI, is now labeled classic, and the Microsoft Learn page for it says it has been replaced by a new version that adds AI apps and agents and says new improvements will not be added to the classic one. Some option names differ between the two. Check which one your tenant shows.

Step 2: label what is sensitive

Sensitivity labels are the lasting control for content, because they travel with the file. Two settings matter for Copilot readiness.

First, make sure SharePoint and OneDrive can read labeled files. Microsoft's page on sensitivity labels for SharePoint and OneDrive says that until you enable the feature, those services cannot process files encrypted by a label, which means co-authoring, eDiscovery, data loss prevention, search and other collaborative features will not work for them. It is enabled in the Purview portal or with a SharePoint Online PowerShell setting. It also has limits, such as labels set to expire user access or use double key encryption, which SharePoint and OneDrive cannot process. Read the limitations list before you design your label scheme.

Second, decide what labels mean to Copilot. Purview Data Loss Prevention has a policy location called Microsoft 365 Copilot and Copilot Chat. With a Content contains sensitivity labels condition and the action Prevent Copilot from processing content, you can keep labeled files and emails out of the response. Microsoft's page says identified items still appear in a response's citations, but Copilot does not use their content. A second rule can block prompts that contain sensitive information types. Two limits matter: you cannot put the sensitive-information-type and sensitivity-label conditions in the same rule, and policy updates can take up to four hours to take effect. The page also says DLP does not scan the contents of files uploaded directly into a prompt, and that some prompt-blocking features were in preview at the time we read it.

Labeling at scale means auto-labeling. DSPM's Protect tab suggests creating an auto-labeling policy when sensitive information is found in unlabeled files. Expect to tune it. Test with a small group and review results before widening, as Microsoft itself advises for new labels, which need time to replicate.

Step 3: use a temporary brake, and know what it is

Sometimes you cannot finish cleanup before the license is in users' hands. Microsoft offers a site-level control for exactly that, and it is easy to misread.

Restricted Content Discovery (RCD) limits discovery of content from chosen SharePoint sites in organization-wide search and Copilot responses, and removes AI entry points such as the Copilot button from those sites. Microsoft describes it as a temporary governance control that gives you time to review and right-size access. Its page was updated in October 2026 and lists the facts that matter.

• It does not change permissions. Users who already have access can still open the content directly.
• It applies to SharePoint sites, not OneDrive.
• It does not remove content from the search index, so Purview features such as eDiscovery and auto-labeling keep working.
• It requires a Microsoft Copilot license and SharePoint Advanced Management. It is enabled in the SharePoint admin center or with PowerShell, and by default only SharePoint administrators manage it.
• Changes take time to propagate. For sites with more than 500,000 items, an update could take more than a week.
• Microsoft advises using it selectively, because heavy use reduces what Copilot can find and weakens its answers.

Suitable candidates are sites with high oversharing risk, sites under permission review, and finance or HR sites during a phased rollout. The key point is in the name: this controls discovery, not access. Someone with a link, or with direct access, still reaches the file. Treat RCD as a bridge to cleanup, not a substitute for it.

Restricted SharePoint Search was the earlier version of this idea, and it has a different status. Its Microsoft page opens with a notice that it is retiring, that new enablement is blocked from July 31, 2026, and that RCD should be used instead. The page also says it was never a security boundary, did not change permissions, was limited to an allowed list of up to 100 sites, and did not guarantee that only allow-listed sites would appear, since recently accessed or directly shared sites could still show up. If you already use it, plan to move to RCD and to real remediation.

Step 4: clean up the permissions

This is the step that removes the risk. Work in order of exposure.

• Anyone links. Review and remove sharing links that give access to anyone who has the link, starting with sites that hold labeled or sensitive content. DSPM's item-level remediation can remove a sharing link, and Microsoft warns to use it sparingly because it can break legitimate access. Tell the owner first.
• Broad groups. Replace Everyone and Everyone except external users on sites and files with named groups. The special-groups report tells you which items are exposed that way.
• Orphaned sites. Give every site at least two owners who are responsible for access. Use the SharePoint site access review feature to ask owners to confirm who should have access.
• Stale content. For content not touched in a long time, decide whether to archive or delete it under a retention policy. DSPM suggests a retention policy for content not accessed for at least 3 years; whether that period fits you is a decision for your records and legal teams.
• Guests. Remove external users who no longer need access.
• Defaults. Make the safe sharing choice the default: set sensitivity labels to apply a default sharing link, restrict sharing at the tenant level where the business allows, and label at creation.

For urgent cases, the data access governance page lists Restricted access control to limit a site to a specific group and the change history report to see recent permission changes. For slower, collaborative cleanup it suggests site access reviews, so owners decide. Choose by the sensitivity of the content, the amount at risk and the likely disruption to users.

Step 5: pilot, then widen

Microsoft's foundational deployment blueprint for a secure and governed Copilot foundation is organized in three pillars: remediate oversharing, set up guardrails and meet regulations. It says SharePoint Advanced Management is included with a Copilot license and pairs it with Purview. A sensible order follows from it. Run the reports. Fix the top exposures. Turn on labels and DLP for the most sensitive content. Use RCD for the sites you cannot finish in time. Then give Copilot to a pilot group, ask them to try the questions a curious employee would ask, such as salary bands, board papers, client matters and patient lists, and see what comes back. Every surprise is a permission problem you can now fix.

Also decide what you will watch afterwards. Purview audit logs record Copilot interactions, and DSPM includes an AI observability view of AI apps and agents. Oversharing returns as people share new things, so put the weekly assessment and the monthly activity reports on someone's calendar.

What readiness does not cover

Fixing oversharing is the first gate, not the only one. Copilot Studio agents, connected data sources and third-party plugins bring their own questions, which we cover in our article on monitoring AI agents in Microsoft 365. Regulated firms also have professional duties around client and patient data. Our articles on the HIPAA Security Rule, ABA Formal Opinion 512 and Florida AI rules cover several of them. None of the controls above promises that no sensitive content will ever surface. They reduce the chance and let you show what you did.

Where a free security assessment fits

The hard part of readiness is knowing what you actually have. Our free security assessment is the place to start. It is a conversation with a security engineer followed by a written summary with your first three priorities, and, if you use or plan to use Microsoft Copilot, it includes a Copilot readiness look at what the assistant could reach and where oversharing needs fixing first. The assessment itself is a review and a report: the permission changes described in this article are separate work that we scope with you before anything starts.

Sources

• Microsoft Learn: Restrict discovery of SharePoint sites and content (Restricted Content Discovery)
• Microsoft Learn: Restricted SharePoint Search (retiring)
• Microsoft Learn: Learn about Microsoft Purview Data Security Posture Management (DSPM)
• Microsoft Learn: Prevent oversharing with data risk assessments from Microsoft Purview DSPM
• Microsoft Learn: Data Security Posture Management for AI (classic)
• Microsoft Learn: Data access governance reports for SharePoint sites
• Microsoft Learn: Microsoft Purview DLP for Microsoft 365 Copilot and Cowork
• Microsoft Learn: Enable sensitivity labels for files in SharePoint and OneDrive
• Microsoft Learn: Secure and governed data foundation for Microsoft Copilot, foundational deployment guidance

This article is general information, not legal advice. It describes Microsoft Learn documentation as read on November 8, 2026. Product names, features, licensing, preview status and limits change often, and some features named here were in preview or rolling out at that time; confirm current terms with Microsoft and your licensing adviser. Nothing here states that any configuration prevents all exposure of sensitive content. Removing sharing links and changing permissions can interrupt legitimate work: test and tell owners first. The free security assessment is a conversation and a written summary; it does not change your environment. Last reviewed November 8, 2026.

Keep reading

Next · SecurityProving what you sent: immutable email archives with Amazon S3 Object Lock and AWS KMSA saved copy of an email is not proof. To show years later what was sent, you need storage nobody can delete and a cryptographic record made at capture time. This article compares Amazon S3 Object Lock modes, AWS KMS options (SSE-KMS only, HMAC keys, asymmetric signing keys), supporting records such as S3 checksums and AWS CloudTrail, how to capture mail with Amazon SES or Microsoft 365 journaling, and how to verify it all later.11 min read

• Local, Private or Frontier? How to Use AI Without Exposing Sensitive Data6 min
• Local AI vs Frontier APIs: Cost, Privacy and Governance of Running Open-Weight Models on Your Own GPUs8 min