If you’ve ever sat through a data-security meeting where someone suggested “let’s just label everything manually,” you know the pain. In theory, it sounds simple. In practice, it’s a mess: thousands of files, inconsistent tags, and a growing risk that something sensitive ends up shared with the wrong person.

Google is trying to fix that with a new feature that quietly slipped into beta this month: Gemini-powered AI classification for Google Drive. Instead of asking employees to tag every document themselves, admins can now tell Gemini what “confidential” or “internal-only” looks like in plain English, and let the model scan Drive and apply labels automatically.

At its core, this is an upgrade to Google’s existing Drive Labels system. For years, Workspace admins have been able to create classification labels like “Public,” “Internal,” “Confidential,” and then either ask users to apply them manually or set up basic rules to auto-apply them in some cases.

The new piece, now in open beta, is that you can use Gemini large language models to inspect file content and decide which label fits, based on custom instructions you write. Think of it as giving your security team a very fast, very tireless intern who can read documents, understand context, and stick the right sticker on each file.

Google describes it as “Gemini-based data classification” in Drive, and it’s rolling out gradually to eligible customers through September 30, 2026. It shows up in the Admin Console under Security > Access and data control > Data classification, where admins define the labels and the instructions Gemini should follow.

How it works

Here’s the practical flow:

• An admin creates or picks a classification label, like “Confidential – HR” or “Restricted – Finance.”
• They write plain-language instructions for Gemini: for example, “Label as ‘Restricted – Finance’ any document that contains unreleased financial results, merger details, or internal forecasts not meant for external sharing.”
• They choose which users’ files should be scanned and which labels should be auto-applied.
• Gemini starts inspecting file content across Drive and automatically applies labels where it thinks the instructions match.
• Admins can monitor how many files were classified and how often users accept or change those labels in the Drive events log.

End users don’t flip a switch for this; they just start seeing more accurate labels on their files, and possibly different sharing behavior depending on how the organization has tied those labels to security policies.

For small startups with a few hundred documents, manual labeling might be annoying but manageable. For mid-size and large organizations, it’s a different story.

Google Drive data classification, in general, is about identifying and tagging sensitive files so they’re protected according to policy: limiting external sharing, triggering warnings, enforcing retention rules, or feeding into Data Loss Prevention (DLP) controls. The problem has always been scale and consistency.

AI-powered classification changes that equation in three ways:

• Scale: It can scan huge volumes of files across formats, including PDFs and images with text (via OCR), far faster than any human team.
• Consistency: Instead of five different people interpreting “confidential” five different ways, you get one set of instructions applied uniformly.
• Context awareness: Unlike older rule-based systems that just look for keywords or patterns, Gemini can understand broader context: the difference between a public press release and an internal draft that mentions the same numbers, for example.

That matters for compliance (GDPR, HIPAA, financial regulations), for internal governance, and for basic “don’t accidentally share the salary sheet with the entire company” hygiene.

Where this fits in Google’s AI security push

Over the past year, Google has been layering more AI into Workspace security and data protection:

• AI classification for Drive has existed in earlier forms, using custom-trained models to identify and label files.
• Admins have been able to tie Drive labels to DLP policies, Vault retention rules, and sharing restrictions, so a label isn’t just metadata—it actually changes how a file can be used.
• More recently, Google has emphasized that its generative AI features, including Gemini in Workspace, are designed to respect existing data protections: if a file is blocked by policy, Gemini shouldn’t be able to pull it into a response.

The new Gemini-based classification is another step in that direction: using the same underlying models people already know from Gemini chat and Workspace AI features, but pointed at the unglamorous, critical job of keeping data organized and protected.

Who can use it (and when)

Not every Workspace customer gets this automatically. According to Google’s documentation, Gemini-based data classification in Drive is included with:

• Enterprise Plus
• Frontline Plus
• Google AI Pro for Education
• And certain add-ons like AI Security or legacy Gemini Enterprise plans

For eligible domains on Rapid Release and Scheduled Release tracks, the feature is rolling out gradually in open beta, with completion targeted by September 30, 2026. Admins will see the configuration options appear in the Admin Console as the rollout reaches their organization.

What changes for employees

From an end-user perspective, the ideal outcome is boring in a good way:

• You create or upload a document as usual.
• Over time, you notice more files have clear sensitivity labels.
• When you try to share something labeled “Restricted,” you might see stronger warnings, need approvals, or be blocked from sharing externally, depending on your company’s setup.

You don’t have to learn a new UI or remember a new tagging system. The heavy lifting—reading content, understanding context, applying labels—happens in the background.

That said, organizations will still need to do the human work up front: deciding what categories matter, writing clear instructions for Gemini, and communicating to staff what those labels mean in practice. AI can automate the tagging, but it can’t decide your company’s risk tolerance or regulatory obligations for you.

What this doesn’t do (yet)

It’s worth tempering expectations. This is beta software, and it’s focused on a specific job:

• It classifies files in Google Drive based on admin-defined instructions; it’s not a full-blown, cross-platform data discovery tool for every cloud service you use.
• It doesn’t replace the need for good security policies, training, or oversight. Admins still need to monitor results, refine instructions, and make sure labels are tied to meaningful controls.
• It’s not a magic “fix everything” button for data leaks. It reduces risk by making sensitive files easier to identify and protect, but human behavior and configuration choices still matter.

Think of it as a powerful new tool in the security toolkit, not a silver bullet.

Why this could be a big deal

If you zoom out, this is one of those quiet infrastructure upgrades that could shape how companies handle data for years. As more work moves into cloud suites like Workspace, the old model of “we’ll train everyone to label everything correctly” becomes less and less realistic.

Gemini-based classification offers a different path: define your categories once, describe them in plain language, and let AI do the repetitive, error-prone work at scale. Over time, that should mean fewer mislabeled files, fewer accidental exposures, and a clearer picture of where sensitive data actually lives.

For IT and security teams, that’s a win. For everyone else, it’s one less thing to worry about when hitting “Share.”

If you’re an admin in an eligible Workspace edition, expect to see this option appear in your Security settings over the next few weeks. For now, it’s in open beta, which means it’s ready for real-world testing but still evolving. The organizations that experiment early—writing good instructions, testing on sample folders, and tying labels to real policies—are likely to get the most value once the feature matures.

In a world where “AI in the enterprise” often means flashy demos and vague promises, this is a refreshingly concrete use case: AI that quietly reads your documents, understands what they are, and helps keep them safe.