NEW · What your AI agents can access vs. what they actually do with it. Live with ISMGSee how →
Last Accessed Property: Microsoft 365 Is Full of Documents No One Remembers

Last Accessed Property: Microsoft 365 Is Full of Documents No One Remembers

Dark data makes up 55% of what organizations store, and Microsoft's lastAccessedDateTime doesn't even cover SharePoint or OneDrive. See what actually finds it, and why Copilot makes it urgent.
Bruno Kurtic, President and CEO, 
Co-founder, Bedrock Data

Bruno Kurtic

President and CEO, Co-founder

November 4, 202611 min read
Bedrock Data logo and text, plus sign, and M365 logo and text.

Key takeaways

  • Dark data, information your organization stores but never uses, makes up 55% of organizational data, per Splunk's State of Dark Data survey.
  • Microsoft Graph's lastAccessedDateTime property does not work for SharePoint Online or OneDrive for Business, a gap most teams find too late.
  • Stale SharePoint, OneDrive, and Teams content is exactly what Microsoft 365 Copilot can resurface to anyone who still holds old access.
  • Purview's auto-labeling and SharePoint Advanced Management give you real discovery tools beyond the Graph API alone.
  • Bedrock Data classifies SharePoint, OneDrive, and Teams content continuously alongside the rest of your data estate, with agentless analysis that keeps raw data in your environment.

Someone on your team needed a file last week and could not find it. They checked three SharePoint sites, a OneDrive account that may belong to someone who left in 2023, and a Teams channel nobody has posted in since the last reorg. They found four versions of the file, none current, all three still shared with people who no longer need access.

That search happens daily across most Microsoft 365 tenants, and it points to a bigger problem than lost time. A typical tenant accumulates thousands of documents nobody opens, reviews, or remembers exist, and some of them hold exactly the sensitive data your compliance team assumes is locked down. This guide covers what that dark data looks like inside Microsoft 365, why it is riskier than it seems, and how to find and fix it.

Dark data is the information your organization stores but never uses or analyzes again, and in Microsoft 365 it piles up in SharePoint libraries, OneDrive accounts, and abandoned Teams channels. Microsoft Graph's lastAccessedDateTime property does not even cover SharePoint Online or OneDrive for Business. The more reliable signals come from Purview Data Lifecycle Management's new last-accessed retention control, Purview auto-labeling policies, and SharePoint Advanced Management's access reviews.

What is dark data in Microsoft 365?

Dark data is information your organization collects and stores but never actively uses, analyzes, or revisits. It is the shadow side of every data inventory: present on disk, absent from any decision. In a Microsoft 365 tenant, it rarely looks abstract. It looks like SharePoint libraries nobody opens, OneDrive folders tied to employees who left years ago, old CSV exports from a project that shipped, and abandoned Teams channels whose last message predates the current org chart.

Dark data overlaps with, but is not identical to, ROT (redundant, obsolete, trivial) content. ROT describes data someone has already reviewed and confirmed has no remaining business value. Dark data is broader: it includes ROT, plus everything nobody has looked at closely enough to classify either way. Splunk's State of Dark Data survey puts dark data at 55% of organizational data industry-wide, a figure large enough that most of it is sitting inside a system like Microsoft 365 by default.

  • Dark data: Stored and unanalyzed; business value unknown.
  • ROT data: Reviewed and confirmed redundant, obsolete, or trivial.
  • Stale permissions: Access grants that outlived the reason they were created.

Why do forgotten files pose hidden risks?

Content becomes dark data gradually. A project wraps, the owner moves teams or leaves, and the SharePoint library or OneDrive folder they left behind keeps its permissions exactly as they were on the last day anyone paid attention to it. Nobody revokes access because nobody remembers to. That stale access is the actual exposure: a file does not need to be opened to be a liability; it only needs to still be reachable by people who should not reach it anymore, or by Microsoft 365 Copilot summarizing it into an answer for whoever asks.

Unknown sensitive data sitting in forgotten libraries is also a compliance problem the moment a regulator or auditor asks what personal or regulated data you hold and where. IBM's 2024 Cost of a Data Breach report found that 35% of breaches involved data sitting in unmanaged sources, and those breaches took 26.2% longer to identify, took 20.2% longer to contain, and cost an average of $5.27 million, well above the overall average. Forgotten Microsoft 365 content is a milder version of the same pattern: data nobody is actively watching.

Take a SharePoint folder of legacy financial reports. It has not been opened in 18 months, yet it is still shared with 27 users across two departments, several of whom have since changed roles. Nobody remembers it is there until it shows up in a discovery request, an audit sample, or a Copilot answer to someone who stumbled onto access years ago and never lost it.

How can you discover dark data across SharePoint, OneDrive, and Teams?

Discovery fails when it is a project instead of a standing process. A once-a-year manual audit is already stale by the time it is published, because new files, shares, and departures happen every day. The three methods below, run continuously and pulled together through the Microsoft Graph APIs, give you a standing picture instead of a once-a-year snapshot.

How do you use last-accessed metadata?

Here is the catch most teams hit: Microsoft Graph's fileSystemInfo.lastAccessedDateTime property is explicitly not available for items in SharePoint Online or OneDrive for Business; it only applies to the per-user recent file list. Microsoft has announced retention based on last access for SharePoint and OneDrive in Purview Data Lifecycle Management, with preview planned for early 2027.

How do automated classification and sensitivity labels help?

Microsoft Purview auto-labeling policies apply sensitivity labels service-side to files already at rest in SharePoint and OneDrive, using sensitive information types and trainable classifiers, up to 100,000 files a day tenant-wide. Monitor coverage in Content Explorer, and pair label data with last-accessed metadata for a simple risk score: a regulated label on a file nobody has opened in a year outranks the same label on an active file. Bedrock Data's AI-driven categorization does this scoring continuously across every data store it sees, not only Microsoft 365, which is the core of effective metadata management at scale.

How do you audit permissions and oversharing?

SharePoint Advanced Management's data access governance reports export sharing exposure directly: sites shared with "everyone except external users," active anonymous or org-wide links, and a full site permissions baseline. Cross-reference the sites that show up there against the stale, unaccessed libraries your discovery pass already flagged; overshared and unaccessed together is the highest-priority combination. Full-Context Entitlement Analysis resolves this down to the identity level, including nested groups, and ranks the riskiest identities with an Impact Score.

How do you build a dark data remediation policy?

Tooling without a policy just generates a longer list of problems nobody owns. Decide, before you scan anything, how your organization will prioritize what it finds, who gets notified, and what happens to content that turns out to be genuine ROT. The three tracks below cover that decision in order.

How do you prioritize by risk and value?

Score every flagged item on two axes: business value (is anyone still using or citing it) and risk (sensitivity label, share count, and access age). High risk and low value—an unaccessed file with a regulated label and broad sharing—gets remediated first. A reasonable starting threshold for Microsoft 365: flag anything unaccessed for 12 months or more that also carries a sensitivity label or an "everyone" link.

How do you automate owner notifications and reviews?

Microsoft's site access reviews let IT delegate up to 100 sites at a time to their owners, with a tailored email describing the specific oversharing issue found. Owners review, remediate, and close the loop with comments that flow back to IT, creating a timestamped, named audit trail that holds up for a compliance review. Access reviews and least-privilege enforcement extend that same workflow beyond SharePoint to every data store with an owner to notify.

Should you archive or delete ROT content?

Check retention labels and legal hold status before touching anything. Content under an active hold or an unexpired retention period cannot be deleted regardless of how old or unused it looks. For everything else, cold storage options like Microsoft 365 Archive move genuinely inactive content out of active storage and, notably, out of the content Copilot can surface, while deletion is reserved for content a records manager has formally confirmed as ROT.

See the platform behind this: If you want to understand how continuous discovery and classification work before you request a demo, start with Bedrock Data's Metadata Lake.

What is Bedrock Data's approach to dark data discovery?

Everything above works as a one-time project or a continuous intelligence layer; Bedrock Data is built for the second option. It runs agentless, analyzes data in place, and covers Microsoft 365 alongside your cloud, SaaS, and on-prem stores, so Microsoft 365 findings sit in the same context as everything else. The two capabilities below are what that looks like day-to-day.

What does continuous scanning and classification look like?

Bedrock Data's Serverless Outpost Architecture deploys ephemeral, agentless compute inside your own AWS account via infrastructure as code, scanning continuously and scaling to zero when idle. Scanning and classification happen inside your environment under Zero Data Access; only metadata reaches Bedrock Data's platform. Every result lands in the Metadata Lake as a continuously updated record of what data exists, where it lives, and whether anyone still touches it.

What can a policy-driven engine actually do?

Bedrock Data's natural-language policy engine reads your existing data policies, flags aged and overshared content that matches them, scores the identities that can still reach it, and notifies owners. It hands the actual unshare, quarantine, or deletion action to your existing tools through its bidirectional integrations, rather than taking that action autonomously. For a plain-language view of exactly who can still reach a flagged file, Bedrock Data's Metadata Lake Copilot answers that as a direct query.

How can you bring dark data into the light?

Dark data in Microsoft 365 is not a one-time cleanup; it is a standing condition of how SharePoint, OneDrive, and Teams get used. The fix is continuous, not heroic: track last-accessed signals through Purview's retention control rather than the Graph property that does not cover SharePoint Online, let auto-labeling do the classification work, run data access governance reports and site access reviews on a schedule, and route anything genuinely obsolete through archiving or deletion with retention checked first. See what that looks like on your own tenant with a Bedrock Data demo.

FAQs

What qualifies as dark data in Microsoft 365?

Any SharePoint, OneDrive, or Teams content your organization stores, but no one actively opens, reviews, or relies on for a business purpose, qualifies as dark data. That includes orphaned project folders, OneDrive accounts tied to former employees, old exports, and Teams channels gone quiet. It overlaps with ROT content once someone confirms a file is genuinely redundant, obsolete, or trivial, but dark data also includes anything nobody has reviewed closely enough to classify either way.

How often should last accessed audits run?

Run discovery continuously rather than as a periodic project; manual, one-off audits are outdated the moment they finish, since new files, shares, and departures happen daily. If continuous scanning is not available yet, treat quarterly as the minimum cadence for most tenants, with a monthly pass for sites already flagged as overshared or holding regulated data under an active retention label.

Does deleting dark data break retention rules?

It can, if you delete before checking for an active retention label, legal hold, or regulatory requirement tied to the content's classification. Some files that look abandoned sit under a hold for litigation or a statutory retention period that has not expired. Always confirm retention and legal hold status first, route anything regulated to archiving instead of deletion, and keep a record of what was reviewed and why.

Can sensitivity labels be applied in bulk to legacy content?

Yes. Microsoft Purview's auto-labeling policies apply sensitivity labels service-side to existing files at rest in SharePoint and OneDrive, not only to new content, up to a combined limit of 100,000 files per day across all active policies in a tenant. For files that predate your current sensitive information types, run on-demand classification first, since standard auto-labeling only evaluates content created or modified after a rule's creation date.

What quick wins reduce dark data risk within 30 days?

Pull a data access governance report to find your most overshared sites, then launch site access reviews on the worst offenders. Turn on auto-labeling for your most common sensitive information types. Flag sites inactive for 12 months or more for an ownership check. None of this requires new tooling, and together they remove the riskiest exposure before a deeper remediation program starts.

Share:

Related Resources