Blackline reviewing a Polish lease agreement: 19 personal details marked by type, with a checklist on the right
Blackline mark
Blackline
Draw the line under private data.
Find and permanently remove personal data from contracts, scans and AI prompts — on the device, in seconds.
FOR LEGAL · COMPLIANCE · SECURITY TEAMS
THE PROBLEM
Every document your team shares still has people in it.
Names, ID numbers, bank accounts, diagnoses — buried in contracts, claims, case files and the prompts people paste into AI.
A discharge summary received by fax: grainy, low resolution
An invoice photographed with a phone: angled, shadowed, blurred
A form filled in by hand in blue pen
A crumpled, coffee-stained employment contract scanned at an angle
An employment contract scanned crooked
FAX · 13
PHONE PHOTO · 10
HANDWRITTEN · 9
CRUMPLED · 15
CROOKED · 12
WHERE IT HIDES
Where search can’t reach.
Faxes, phone photos, handwriting, crumpled scans. No text layer, nothing to search — and nobody reads every line.
59
personal details on these five pages alone
THE COST OF GETTING IT WRONG
One missed line is all it takes.
4%
of global turnover
or €20 million, whichever is higher: the GDPR ceiling for the most serious breaches.
1 paste
into an AI chat
and a client’s name, ID and account have left the company — with no record of where they went.
≠ gone
A black box isn’t redaction
Drawn over a PDF, it often leaves the words underneath — selectable, searchable, copyable.
Fine ceiling: GDPR Article 83(5).
MEET BLACKLINE
Blackline finds it. You confirm it. It’s gone.
Automatic detection, a human sign-off and true removal — without the document ever leaving the device.
LIVE · A REAL LEASE AGREEMENT
Watch a contract go clean.
web embed
BEFORE · AFTER
Not covered. Removed.
BEFORE
Original employment contract showing the employer, the employee’s name, addresses, salary, IBAN, date of birth, NI number, phone and email
AFTER
The same contract after Blackline: every personal detail replaced by a solid black bar
Copy the text out of the saved file: the names, numbers and accounts are not there. Real output from a sample contract.
SCANS & PHOTOS
Crooked, crumpled, photographed — still caught.
Before: a crumpled, coffee-stained employment contract scanned 5.5 degrees off, with handwritten notes
BEFORE
Screen recording: the crumpled scan is dropped into Blackline, straightened, 28 items are marked, then Preview shows them blacked out and the copy is saved
Straightened 2.8°, read, 28 items marked and removed. Also tested: crooked scan 23 · phone photo 20 · handwriting 13.
BROWSER GUARDRAIL · AI CHATS
Paste into any AI chat. Only the safe version lands.
web embed
ONE ENGINE, EVERYWHERE
Runs where your data runs.
Laptop
Desktop app
READY TODAY
macOS and Windows. Documents, scans, whole folders — fully offline.
Browser window
Browser guardrail
READY TODAY
Every paste into any web app, AI chat or form — cleaned first.
Server rack
On-prem server
SET UP WITH YOUR IT
One engine for a whole office, inside your network. No data leaves it.
Workflow of connected steps
Pipelines & automations
SET UP WITH YOUR IT
n8n, Make, Airflow or your own scripts call one API before data moves on.
POST /api/anonymize {"text": "Call Jan Kowalski on +48 601 234 567", "mode": "pseudonymize"}
→ "Call [PERSON_1] on [PHONE_1]" · /api/deanonymize puts the real values back after the AI step.
WHAT’S INSIDE
Built for the documents you actually have.
Every format
PDF, Word, text, photos and scans.
24 EU languages
Even mixed in one document.
35 kinds of data
From names to health details.
Checked IDs
PESEL, NIP, IBAN and cards are verified by check digit.
Hidden text found
Hidden layers and off-page text in PDFs come to light.
Whole folders
6 files, 123 items found, one Save all.
Fully offline
No cloud, no API keys, no telemetry.
Your house rules
Always keep, always remove — one list, shared as CSV.
PROOF
Measured head-to-head against the standard open-source setup.
1.00
F1 accuracy
vs 0.72 for the baseline
0
false alarms
vs 12 innocent numbers blacked out
0
missed details
vs 3 that would have shipped
<80
milliseconds per paste
1–2k characters, on a laptop CPU
Internal benchmark: 16 documents up to 29k characters; precision on the 9 fully annotated ones. Baseline: Microsoft Presidio + spaCy pl_core_news_md.
TRUST
Built for health, identity and financial data.
Nothing leaves
Runs entirely on the device. No cloud, no telemetry.
Removes, then checks
Text is deleted from the file, then verified before saving.
Sees the hidden
Hidden layers, off-page text and form fields are cleared.
Locked to the app
The engine answers only the app that started it.
Reviewed, nothing open
34 security findings: 31 fixed and tested, 3 settled by decision.
Air-gap ready
After one model download, it never needs a network.
WHERE IT PAYS OFF
Every team that shares documents.
Legal
Disclosure bundles, subject-access requests, case files for outside counsel.
HR & recruiting
CVs, contracts and employee records shared with managers and agencies.
Healthcare & insurance
Claims, referrals and discharge summaries sent to reviewers and partners.
Finance & KYC
Invoices, statements and ID documents for auditors and vendors.
Public sector
Freedom-of-information responses and case records released to the public.
Anyone using AI
Support replies, sales drafts and summaries — without the client inside.
NEXT STEP · 30-DAY PILOT
Run it on your own documents.
WEEK 1
Install
Desktop app and browser guardrail for one team.
WEEKS 2–3
Use it for real
Contracts, claims, case files and AI prompts.
WEEK 4
Measure
What was found, what was missed, hours saved.
Pick a team. Pick a folder. Nothing leaves your building.