Where does your document text actually go?
SafeRedact says files stay in your browser. They do. But the text extracted from them is sent to an outside AI service for detection: Anthropic's Claude API, or AWS in the UK and EU. PII Anomalyzer detects, anonymizes, and redacts PII entirely on your device.
Two different answers to the same question
Both tools redact PII. They get there by different paths. The path matters when your documents are privileged, regulated, or confidential.
SafeRedact
Browser-based, cloud-AI detection
- 1. PDF opens in your browser
- 2. Text is extracted client-side
- 3. Extracted text is sent over TLS 1.3
- 4. Anthropic's Claude API detects the PII (EU route: AWS in the UK and EU)
- 5. Detection results come back to the browser
- 6. Redactions applied in your browser
Original file binary never leaves the browser. Extracted text content does. Source: SafeRedact privacy policy, security page and Trust Center, October 2026.
PII Anomalyzer
Desktop app, local AI detection
- 1. PDF opens on your device
- 2. AI models run on your device
- 3. Detection completes on your device
- 4. Redactions applied on your device
- 5. Output saved on your device
- 6. No third-party API in the path
Document content never crosses your network. See our security page for the full network disclosure.
$99 a year, one document at a time. $249 a year, whole batches.
SafeRedact Personal is inexpensive for a one-off PDF. Bulk, Word, and Excel are an enterprise quote. The question is whether the architecture fits your work.
SafeRedact
$99/yr
Personal: $29/mo, or a $12 day pass
One PDF or image at a time in the browser, with detection routed through Anthropic's Claude API. Free outputs carry a watermark. Enterprise DSAR cases are quoted per case.
PII Anomalyzer
$249/yr
7-day free trial included
Unlimited documents and batches. PDF, Word, Excel, and scans. Nothing sent to a third-party AI.
What each tool actually does
Verified against SafeRedact's pricing, privacy, security, and Trust Center pages as of October 2026.
| Capability | PII Anomalyzer | SafeRedact |
|---|---|---|
| Document text stays on your machine SafeRedact: extracted text is sent to Anthropic's Claude API (EU route: AWS in the UK and EU) | Yes | No |
| Local AI (no third-party API for detection) SafeRedact: detection runs on Anthropic's Claude API, or on AWS for the EU route | Yes | No |
| Detection and redaction work offline SafeRedact: detection needs an internet connection | Yes | No |
| PDF redaction | Yes | Yes |
| Word and Excel files on a self-serve plan SafeRedact: Personal accepts PDF, JPG and PNG only; Word, Excel and email files are in the enterprise case product, quoted per case | Yes | No |
| HIPAA-aligned workflows SafeRedact: vendor states not HIPAA compliant and offers no BAAs | Yes | No |
| Batch processing on a self-serve plan SafeRedact: Personal handles one document at a time; bulk processing is the enterprise case product, quoted per case | Yes | No |
| Email and mailbox files (EML, MSG, PST) SafeRedact: enterprise case product, built for data subject access requests, with an audit trail | No | Yes |
| Re-identification workflow (anonymize, share, translate back) | Yes | No |
| Full-featured free trial (no watermark) SafeRedact: free tier outputs include watermark | Yes | No |
Three contexts where the architecture matters
If any of these apply, sending document text to Anthropic isn't a feature comparison — it's a workflow blocker.
Cloud-AI prohibition
Government contracts, defense work, financial services WISP requirements, internal policies that prohibit third-party AI processing of work product.
The blocker
Routing extracted document text to Anthropic's API, or to AWS on the EU route, is a no-go regardless of encryption in transit. The text leaves your environment to a third-party processor.
In PII Anomalyzer
PII Anomalyzer's local AI keeps document text on your device. No third-party API in the document path means no contract or policy review of an AI processor — there isn't one.
Attorney-client privileged work
Discovery productions, privilege reviews, settlement drafts, client communications under ABA Model Rule 1.6 confidentiality obligations.
The blocker
ABA Formal Opinion 512 (issued July 2024) addresses lawyers' duty to evaluate whether generative AI tools preserve client confidentiality under Model Rule 1.6. Sending privileged text to an outside API at minimum requires that evaluation; keeping the text on your own machine makes it a much shorter one.
In PII Anomalyzer
Local AI means no third-party processor holds your document text, even momentarily. The confidentiality analysis is simpler when nothing leaves your device.
HIPAA-touched documents
Patient records, billing documents with PHI, healthcare disclosure requests, medical research data, any document touched by 45 CFR Part 164.
The blocker
SafeRedact's own security page states: "SafeRedact is not HIPAA compliant and does not offer BAAs." If you handle PHI, SafeRedact directs you to consult your compliance officer.
In PII Anomalyzer
PII Anomalyzer's local processing supports HIPAA-aligned workflows because no document data leaves your device, so no PHI is disclosed to a third-party vendor.
SafeRedact is well-built. Their disclosure is clear.
SafeRedact's privacy policy names Anthropic as a sub-processor. Their security page describes the Claude API routing and the 30-day deletion in plain language. The "files stay in your browser" claim is accurate for original file binaries.
For users who don't have policies prohibiting third-party AI processing, who aren't handling privileged or PHI documents, and who need quick browser-based redaction of a single PDF or image, SafeRedact Personal is a credible option. Their main product is now bulk redaction for organizations answering data subject access requests under GDPR, UK GDPR, CCPA, and similar laws, with an audit trail built for that job; if that is your need, it is purpose-built for it. Their honesty about what they do is a feature, not a flaw.
The difference is architecture. SafeRedact ships your document text to Anthropic (or to AWS in the UK and EU on its EU route). PII Anomalyzer ships the AI to your computer. The right choice depends on whether your threat model includes "outside parties processing your document content."
PII Anomalyzer supports compliance workflows under HIPAA, GDPR, CCPA, FERPA, GLBA, and the EU AI Act. Because no document text reaches a third-party processor, there's no sub-processor to disclose and no BAA gap to work around.
Try local-AI detection, anonymization, and redaction free for 7 days
Run PII Anomalyzer on your own documents — including the ones you can't send to a third-party AI service.
7-day free trial · $249/year · Windows & macOS