David Mitchell, an employee at Acme Corp, has requested his data: 15 documents — emails, Teams chats, payroll, HR records — naming 15+ people. Produce everything about David, the one to preserve, and redact everyone else.
1Load the data subject — the person whose details must be preserved.
2Load the sample pack — emails, Teams chats, spreadsheets and logs.
3Process. The engine flags third-party PII and leaves David’s intact.
4Review each detection, then export the redacted files.
Fictional data, no DPA needed · sample files run the production pipeline · your own documents: support@saferedact.app
SafeRedact fills this form, processes the 15 files, and hands it back to you at review.
Data Subject Details
Enter the data subject's information. Their PII will be preserved — everyone else's will be redacted.
Other ways the same person appears — nicknames, initials, email variants, employee IDs. These are preserved, not redacted. Specific identifiers (emails, IDs, full names) are always preserved; short ones (a bare first name or initials) are preserved where the document clearly refers to the subject, and flagged for your review otherwise.
The AI will preserve all PII matching this individual and redact third-party PII. You'll review all detections before export.
Job settings
Auto-suggested from today's date.
Hashed and sequenced strip identifying text from filenames.
Merging combines the redacted PDFs into a single file for the data subject. Redactions stay flattened either way. Internal-audit reports are always separate.
You'll receive:
Redacted source files, flattened PDFs, detection report, audit trail, and a subject preservation summary.
MinimumBalancedMaximum
Minimum redactionHigher precision. Fewer detections surfaced. Best when the deliverable needs to remain readable for legitimate downstream use.
BalancedRecommended for most DSARs. Catches most third-party PII with manageable review burden.
Maximum redaction (default)Higher recall. The engine flags more candidates for your review team to approve or reject. Best for high-stakes work where defensibility outweighs review burden.
Upload Files
Drag and drop files or a ZIP archive. Supports PDF, EML, MSG, PST, DOCX, XLSX, TXT, HTML, CSV, ZIP.
Drop files or a ZIP here, or click to browse
PDF, EML, MSG, PST, DOCX, XLSX, TXT, HTML, CSV, ZIP
Reading files…
Extracting PST…
Starting…
Extracting PSTStarting…
0 messages extractedCalculating…
Confirm & Process
Review your batch before processing begins.
Processing files…
Extracting text and detecting PII across all files.
0%
Files: 0 / 0Detections: 0Elapsed: 0s
Processing Log
Confidence Dashboard
Review detection confidence before approving results. Files below threshold are flagged for manual review.
Overall Confidence
—
Waiting for data...
Confidence Tiers
High ≥ 90%—
0 detections across 0 files
Medium 70–89%—
0 detections across 0 files
Low < 70%—
0 detections across 0 files
Detection Sources
PII Types Detected
Flagged for Manual Review
0 files
Files with low-confidence detections, zero detections, or processing errors that should be reviewed by your privacy team before sending.
No files flagged — all detections are high confidence.
Audit Trail
Your review and sign-off are required. Final accuracy is your responsibility as the data controller, not SafeRedact's.
SafeRedact flags potential PII as candidates. The left pane shows detections highlighted in context — click any highlight to toggle it on or off. The right pane shows the redacted output in real time. Review every file and add manual redactions for anything the AI missed.
By downloading, you confirm you have reviewed every detection in every file and that the redactions reflect your team's deliberate decisions about what to redact and what to preserve. No redaction is applied without your approval.
Power-user exports
Detection report (CSV)
Every detection as a row — for QA or analysis
Review decision log (CSV)
Every reviewer action with timestamps — the per-event defensibility record
Detection data (JSON)
Full structured data — for programmatic re-import
One decision per entity cascades to every instance in every file
Select a file to preview
Click to toggle:
Red redact
Amber from other file
Green subject (keep)
Dashed skip
Dotted possible subject — confirm
Detection Review
Select a file from the sidebar to preview
Redacted Output
Redacted preview will appear here
Loading entities…
Select an entity from the list
My Cases
Saved on this device in your browser — never uploaded. Open a case, then re-attach its source files in Step 2 to resume without reprocessing.
Your redacted deliverable is ready
Review is complete. Complete payment to release the encrypted download.
Files processed—
PII detections—
Pricing tier—
Per file—
Total—
Included in your download
Redacted source files (originals untouched on your device)
Flattened PDF copy of every file with burned-in redactions
Per-detection report (CSV) — every decision, with confidence scores
Audit trail (CSV) — who toggled what, when
Subject preservation summary
Send feedback
SafeRedact
Enterprise Pilot
Overview
Welcome to your SafeRedact evaluation
SafeRedact is a document redaction platform built for compliance teams handling DSARs, regulatory production, and contract review. This pilot gives your team a controlled environment to evaluate detection accuracy, review workflows, and audit capabilities.
You can evaluate using your own documents or start with our pre-built sample pack — sample files designed to demonstrate detection across common document types and PII patterns. The sample pack contains no real personal data.
Two-Layer Detection
Deterministic pattern matching for structured IDs, plus an AI layer that reads document context to identify sensitive data patterns alone would miss.
Human-in-the-Loop Review
Every detection is surfaced for human verification. AI assists — your team approves. No redaction is applied without review.
Browser-Based Processing
Your files never leave your browser — parsing, review, and redaction run on your device. Only extracted text is transmitted for detection analysis under the DPA's terms — never used for model training, deleted by the AI sub-processor within 30 days; no document content is ever stored.
Audit Trail
Every detection, classification, and review decision is logged. Confidence scoring and PII type distribution are available for compliance reporting.
Pilot Terms
What's included
Your evaluation provides full access to SafeRedact's detection and review pipeline, running on the synthetic sample pack — the same engine, workflows, and audit surface as production.
Duration
30 days
File processing limit
Unlimited during pilot
Supported formats
PDF, DOCX, XLSX, EML, MSG, PST, TXT, HTML, CSV, ZIP
Detection engine
Full pipeline (regex + AI)
Review workflow
Full access
Confidence dashboard
Full access
Audit trail preview
Full access
Redacted file download
Enterprise license
Evaluation data: This evaluation runs on the synthetic sample pack. The data describes no real person, so no data processing agreement is required.
Process
How the evaluation works
The pilot follows the same workflow your team would use in production. There are no training steps or configuration requirements — upload documents and the detection engine runs immediately.
1
Upload documents
Drag files or a ZIP archive into the workspace. Every supported file type is processed automatically. For DSAR workflows, enter the data subject's details to preserve their information while redacting third-party PII.
2
Detection runs automatically
Layer 1 (regex) identifies structured identifiers — SSNs, IBANs, sort codes, passport numbers. Layer 2 (AI) reads document context to find names, addresses, phone numbers, financial data, and other PII that patterns alone would miss.
3
Review and verify
Each detection is highlighted in the document with a confidence score. Click to approve, dismiss, or reclassify. The split-pane view shows the original text alongside the redacted output in real time.
4
Assess results
The confidence dashboard breaks down detections by type and confidence tier. The audit trail logs every decision for compliance review. Use these to evaluate detection accuracy against your internal benchmarks.
Evaluation Criteria
What to look for
A meaningful evaluation requires testing against your actual document types and compliance requirements. We recommend focusing on three areas:
Detection accuracy
Upload documents where you already know the PII locations. Compare SafeRedact's detections against your ground truth. Note any false positives (flagged incorrectly) and false negatives (missed entirely). Context-dependent PII — names mentioned in passing, addresses embedded in prose — is where the AI layer differentiates from pattern-only tools.
Review efficiency
Time how long it takes to review a batch. Compare against your current manual process. The goal is not zero review — it's dramatically less review with higher confidence.
Format coverage
Test across your actual file mix. PDF contracts, Word memos, email chains (EML/MSG), spreadsheets. Note any formats where detection quality varies.
Sample data available: If you'd prefer to start with sample documents before using your own files, download the sample pack.
Ready
Begin your evaluation
Your pilot environment is ready. You have 30 days and unlimited file processing to evaluate SafeRedact against your compliance requirements. When you're ready to discuss production deployment and enterprise licensing, we're here.
These files contain detections that haven't been individually decided — either via the "By entity" or "By file" review. You can still download, but the audit trail will record these files as not fully attested.
As the data controller, you remain responsible for the accuracy of redactions regardless of attestation status.