Tenet·Feature Guides 05 · DLP & Pseudonymization ← All guides
Guide 05 · Data Privacy

DLP & Pseudonymization

A four-phase pipeline that strips private student data out of prompts and uploaded files before anything leaves the device, including the standout trick: swapping real student names for synthetic ones so the AI vendor never sees who is who.

Regex PII redaction: Basic Roster-aware + pseudonymization: Pro
↓ Download the one-pager (PDF)
What it is

Data Loss Prevention, built for AI prompts

DLP stands for Data Loss Prevention. The risk is simple and constant: a student or teacher pastes something sensitive into an AI tool, a roster, a grade book, an IEP, a parent's phone number, and now that data sits in a third party's cloud. Tenet cleans prompts and uploaded files on the device, before they reach the AI vendor, so the sensitive parts never leave the browser.

How it works

The four-phase pipeline

1 · Regex PII
emails, phones, SSNs, DOB, addresses, IDs
2 · Roster PII
your actual student names, even partial
3 · Pseudonymization
real names swapped for synthetic ones
4 · Context classifier
a peer named Sarah vs Sarah the place

Phase 1, Regex PII (BASIC)

Pattern-based removal of the obvious identifiers: emails, phone numbers, Social Security numbers, dates of birth, addresses, credit card and routing numbers, and district-defined student ID formats. This is free in Basic and covers both chat and files.

Phase 2, Roster PII (PRO)

Because Pro knows the district's actual roster, it can catch a real student's name even when there is no pattern to match, for example “Sarah W.” in the middle of a sentence. Patterns alone would miss that; a roster does not.

Phase 3, Pseudonymization (PRO)

Instead of just blanking names out, Tenet swaps each real name for a believable synthetic one before the prompt goes out, then restores the real name locally when the AI replies. The conversation stays natural and the vendor never learns who the student is.

Phase 4, Context classifier (PRO)

Tells the difference between “Sarah went to Paris,” which is a peer, and “Sarah, the capital of,” which is not a person. This cuts false positives so legitimate work is not mangled.

The standout demo

Pseudonymization, step by step

The AI vendor never sees the real name. The student never notices the swap.

Student types: “Help Sarah Wilson revise her essay.”

Tenet sends the AI: “Help Jordan Maple revise her essay.”

AI replies about Jordan Maple.

Student sees: the answer with Sarah Wilson restored on screen.

The synthetic name is gender-coherent so the reply still reads naturally. The map from real to synthetic lives only in that browser tab's memory and is destroyed when the tab closes. No competitor does this.

Coverage

It is not just typed text

The same scrubbing runs on uploaded files, which is where a lot of the real risk lives, because a single file can contain an entire class roster.

Documents

PDF, Word, Excel, PowerPoint, RTF, CSV, and plain text are parsed and scrubbed in the browser before upload.

Images

On-device text recognition reads text inside images, so a screenshot of a roster is scrubbed too.

Structured data

Spreadsheets and CSVs get cell-aware treatment, so names and identifiers in a grade book are caught.

Basic vs Pro

Where the line is

Basic, free
  • Regex PII redaction in chat and all file formats
  • Image text recognition for uploaded screenshots
  • Teacher DLP warnings to prevent accidental sharing
Pro
  • Roster-aware name detection (real student names, even partial)
  • Session-scoped name pseudonymization
  • Context classifier to cut false positives
  • Per-vendor handling so policy can vary by AI tool
Who it sells to

Lead with the right person

Director of IT

The FERPA conversation gets concrete: student PII is removed before it ever reaches an AI vendor, on the device, and the pseudonym mapping is never persisted.

Superintendent

The defensible answer to “are we leaking student data to ChatGPT?” is no, and here is the mechanism.

Teacher

A safety net for the honest mistake of pasting a roster or a grade into AI. Tenet catches it so the teacher does not have to.

Counselor

The student-record classifier flags a student who tries to share IEP, 504, or discipline details with AI.

Common questions

FAQ

Where is the real-to-fake name mapping stored?
Only in the browser tab's memory, and only while the tab is open. When the tab closes, the mapping is gone. It is never written to our servers.
Does pseudonymization make the AI's answer confusing?
No. The synthetic names are coherent, and the real names are restored on screen, so the student reads a normal, useful reply.
What is free versus paid?
Regex PII redaction across chat and files is free in Basic. Roster-aware detection, pseudonymization, and the context classifier are Pro, because they depend on the district roster.
Can it catch a name with no obvious pattern?
Yes, in Pro, because it knows the real roster. That is the difference between pattern matching and roster awareness.
Honest limits

Say this before they ask

Where to set expectations

  • No DLP catches 100 percent of every phrasing. Regex handles structured identifiers well; roster awareness and the context classifier (Pro) raise the catch rate on names. It is strong protection, not a guarantee.
  • Roster-aware features need a roster. Basic is roster-free, so it provides regex redaction only.
  • Coverage is the supported AI platforms in Chrome.
Keep reading

Related guides