Data Subject Access Requests under GDPR require organisations to disclose all personal data they hold about an individual. Before those documents can be released, every piece of third-party personal information — names, email addresses, phone numbers, addresses, national insurance numbers, bank details — must be permanently redacted. This page explains how SafeRedact automates that redaction step, what it costs, how it compares to alternatives, and when you should use it.
The scale problem with DSAR redaction
DSAR volume is increasing 40–60% annually across most industries. What was ten requests per year for a mid-size organisation in 2020 is now hundreds per month for large enterprises. Every major privacy regulation — UK GDPR, EU GDPR, CCPA, and the growing patchwork of US state privacy laws — grants individuals the right to access their personal data, and public awareness of that right continues to grow. Regulatory enforcement has sharpened too: the ICO and European data protection authorities have issued significant fines for late, incomplete, or improperly redacted DSAR responses.
A single DSAR can involve 20,000 or more files. When a data subject requests everything an organisation holds about them, the response package typically includes emails (EML and MSG files), Microsoft Teams chat transcripts exported from Purview eDiscovery, spreadsheets with mixed data in individual cells, PDF reports, Word documents, and every attachment nested inside those emails. One person — often a DPO or privacy analyst — is typically responsible for processing the entire package, with a 30-day statutory deadline under GDPR or a 45-day deadline under CCPA.
The file types are the hard part. EML files contain nested quoted replies where third-party PII appears across multiple layers of conversation history. MSG files are binary and require parsing before content is even readable. HTML exports from Microsoft Teams and Purview eDiscovery have their own structure. CSV files contain mixed data types in single cells — a single cell might include a name, address, and phone number together. Each format requires different extraction and detection logic. Manual review of 20,000 files at this level of complexity is not feasible within any reasonable deadline.
Why visual masking is not redaction
PDF overlay redaction — placing black boxes over sensitive text — is not permanent redaction. The text content underneath the visual overlay remains in the file and can be recovered using copy-paste, standard text extraction tools, or by removing the annotation layer in any PDF editor. ICO enforcement actions have cited recoverable redaction as a compliance failure, and several high-profile data breaches have resulted from organisations releasing documents with visual-only redaction that recipients were able to reverse. If the underlying text is still present in the file, it has not been redacted.
Permanent redaction requires removing the text content itself, not masking it visually. SafeRedact produces plaintext output with [REDACTED_TYPE] markers — for example, [REDACTED_NAME], [REDACTED_EMAIL], [REDACTED_NINO]. The original PII is deleted from the output, not hidden behind a visual element. There is no annotation layer to remove, no hidden text to extract, and no metadata to recover. The redaction is irreversible by design.
How SafeRedact works
1. Upload your eDiscovery export
Drag and drop a Purview content search export, a ZIP of individual files, or select files manually. SafeRedact supports EML, MSG, HTML (Teams transcripts), CSV, TXT, PDF, and DOCX. Nested ZIPs are extracted automatically. Files are processed in your browser — document content is not uploaded to any server until the AI detection step.
2. AI detection across all files
SafeRedact uses a two-layer detection pipeline. Layer 1 is deterministic regex matching for structured identifiers: National Insurance numbers, email addresses, phone numbers, postcodes, sort codes, and bank account numbers. Layer 2 is contextual AI analysis using Anthropic's Claude, which identifies names, addresses, salaries, dates of birth, and other PII that requires understanding context. In DSAR mode, the pipeline preserves the data subject's information and redacts everyone else's. Processing runs at 8.8 files per minute. A 100-file pilot completes in under 12 minutes. A 5,000-file batch runs overnight.
3. Review every detection
Every AI detection is presented for human review in a split-pane interface. The left panel shows the original document with PII highlighted — yellow for third-party data that will be redacted, green for the data subject's data that will be preserved. The right panel shows the redacted output with [REDACTED_TYPE] markers. Click any detection to see its type, confidence score, and source (regex or AI). Accept or reject each detection individually. Add manual redactions for anything the AI missed by selecting text and choosing a PII type. When you are satisfied with every file, enter your name and role and sign off.
4. Export and pay
Download the complete package: redacted files with [REDACTED_TYPE] markers, an audit trail CSV listing every detection with its accept/reject status, and a compliance report PDF with the reviewer's sign-off ready to attach to your DSAR response. Payment is collected at this point — after review, not before.
Pay after review, not before
Every other redaction tool charges before you see results. You pay a subscription fee, upload your files, and hope the output is usable. If the detection quality is poor or the tool does not support your file types, you have already paid.
SafeRedact charges after review. You upload, process, and review every detection — adding manual redactions where needed — before any payment is required. A 100-file pilot is completely free with no credit card required. You pay only when you have confirmed the output is ready for release.
100-file pilot: completely free, no credit card. Process, review, and export before deciding.
How SafeRedact compares
| Feature | SafeRedact | Manual Process | OneTrust / Osano | Adobe Acrobat |
|---|---|---|---|---|
| Handles eDiscovery exports (EML, MSG, Teams) | ✓ | ✗ | ✗ | ✗ |
| AI PII detection | ✓ | ✗ | ✗ | ✗ |
| Manual redaction override | ✓ | ✓ | ✗ | ✓ |
| Reviewer sign-off with audit trail | ✓ | ✗ | ✓ | ✗ |
| Pay after review | ✓ | N/A | ✗ | ✗ |
| 100-file free pilot | ✓ | N/A | ✗ | ✗ |
| Per-job pricing (no subscription required) | ✓ | N/A | ✗ | ✗ |
| DPA included | ✓ | N/A | ✓ | ✗ |
OneTrust and Osano manage the DSAR workflow — intake, tracking, deadlines, deletion confirmation. SafeRedact handles the document redaction step that those platforms do not perform. They are complementary tools, not competing ones. If you use OneTrust or Osano for DSAR management, SafeRedact is the redaction layer that sits between document collection and disclosure.
Pricing
- Pilot: 100 files free — no credit card required
- Starter: up to 1,000 files — $499 per job
- Standard: 1,001–10,000 files — $0.35 per file
- Enterprise: 10,000+ files — $0.20 per file
Payment is required only after review and sign-off. There is no subscription and no minimum commitment. Each job is priced independently.
See the enterprise pilot guide →
Compliance infrastructure
SafeRedact is built on SOC 2 Type II certified infrastructure. Anthropic (AI processing), Vercel (hosting), and Supabase (authentication and metadata) are each independently certified. No document content is stored on SafeRedact servers after processing. Only job metadata — file names, detection counts, and audit logs — is retained.
A Data Processing Addendum is included with every account, incorporated into the Terms of Service on signup. Separately executed DPA documents for UK GDPR, EU GDPR, and CCPA are available at /enterprise/legal for enterprise customers.
Frequently asked questions
What file types does SafeRedact support for DSAR redaction?
SafeRedact supports EML, MSG, HTML (including Microsoft Teams transcripts exported from Purview eDiscovery), CSV, TXT, PDF, and DOCX. Purview content search exports are natively supported — drag and drop the exported ZIP and SafeRedact extracts and processes every file.
How long does a 20,000-file DSAR take to process?
At 8.8 files per minute with full AI detection, a 20,000-file batch takes approximately 38 hours of unattended processing. A 100-file pilot completes in under 12 minutes. Processing runs in the background — you do not need to keep the browser tab open for the entire duration.
What is the difference between SafeRedact and OneTrust for DSARs?
OneTrust manages the DSAR workflow: intake, identity verification, tracking, deadline management, and deletion confirmation. SafeRedact handles the document redaction step — identifying and removing third-party PII from the files before you disclose them to the data subject. Most organisations that process significant DSAR volumes use both a workflow tool and a redaction tool.
Is there a Data Processing Agreement?
Yes. A Data Processing Addendum is included with every SafeRedact account, incorporated into the Terms of Service on signup. Enterprise customers who require a separately executed DPA can download jurisdiction-specific versions (UK GDPR, EU GDPR, CCPA) from /enterprise/legal.
Does SafeRedact store document content?
No. Documents are processed in memory. Only job metadata (file names, detection counts, processing timestamps, and audit logs) is retained for audit trail purposes. Anthropic's no-retention API policy applies to all AI processing — text sent for PII detection is not retained or used for model training.
When do I pay?
After review. You upload files, run AI detection, review every detection in the split-pane interface, add manual redactions for anything the AI missed, and sign off with your name and role. Payment is collected only at the point of export — after you have confirmed the redacted output is ready for disclosure.