WorkGPT
Rows of wooden catalog drawers, each with a white index label
Photo: Photo by cottonbro studio on Pexels

Google Drive can now label sensitive files with Gemini from plain instructions

Google made Gemini-powered data classification in Google Drive generally available on 17 November. Admins describe in plain words what counts as, for example, confidential, and Gemini reads files and applies the right label. Before, admins had to pick and label training files by hand so a model could learn.

Two new tools come with the release. Admins can test their instructions in the Admin console and see which label Gemini would give before turning it on. They also get near real-time reports: how many files were labeled, how often users accepted the label, and a Drive inventory.

People who own or edit a labeled file can accept or change the label, and the audit log records it. Gemini does not overwrite labels that people or data loss prevention rules set. Turning on auto-apply starts a one-time scan of older files, which can take weeks or months. Files need at least about 50 words of readable text.

The feature is for Enterprise Plus, Frontline Plus and Google AI Pro for Education. The rollout runs until 15 December.

What it means for your work

Labels are what make sharing rules and retention rules work. If most files have no label, those rules protect very little. This makes labeling a large Drive much faster.

Google also points to AI agents: labels can stop agents from reading sensitive files. Test the instructions on one shared drive first, and check the acceptance numbers before you widen it. For related help with Gemini at work, see Gemini skills that read Drive files.

Sources