Redact PII from AI Prompts, Files & DocumentsBefore ChatGPT, Claude or Gemini See It

Yes. Paste the AI's response back into the tool and click Reveal. The 1-Click Reveal engine swaps every token ([NAME_1], [EMAIL_2], etc.) back to the original value using your volatile local session map — zero server round-trips, zero database storage.

AI Tokens mode replaces sensitive entities with structured badges like [NAME_1] and [EMAIL_1], preserving contextual roles for LLMs (ChatGPT, Claude) and allowing 1-click restoration via Reveal Originals. Blackout mode applies permanent, solid black raster bars over sensitive pixel coordinates, physically destroying vector text layers for court filings, legal discovery, and FOIA releases with zero underlying text leakage.

Most conventional PDF editors and markup tools merely overlay black rectangles as a visual layer, leaving underlying OCR text, character coordinates, and metadata intact and extractable via Cmd+A copy or pdftotext. PrivacyScrubber executes true Secure Raster Flattening: pages are re-rendered in local RAM, PII coordinates are overwritten at the pixel level, and a fresh PDF is synthesized with zero underlying text layers, guaranteeing 100% leak-proof redactions.

Supported formats include .txt, .docx, .pdf (text-layer and scanned images via Tesseract OCR), .xlsx, and .csv. With Batch File Redaction and Excel Spreadsheet Scrub, you can download sanitized files with PII removed or export redacted PDFs with permanent raster flattening (zero underlying text layers).

Yes. Most conventional redaction tools only mask visible text or cell contents, leaving hidden document properties intact. Microsoft Word (.docx) and Excel (.xlsx) files are compressed XML archives that store author identity (<dc:creator>), modifying users (<cp:lastModifiedBy>), corporate affiliation (<Company>), and revision timestamps inside docProps/core.xml, docProps/app.xml, and workbook properties. PrivacyScrubber's in-RAM sanitizer intercepts these archives in local browser memory, wipes original author and organization tags, eliminates custom metadata, and overwrites properties with anonymous synthetic signatures (PrivacyScrubber (Zero-Trust)) before export. Your exported files are completely clean both visibly and metadata-wise, with zero network transmission.

When users upload or drop documents (such as resumes or legal agreements) into ChatGPT or Claude, PrivacyScrubber intercepts the file in local browser RAM. Binary formats (.pdf, .docx) contain hidden metadata, revision histories, and XML streams that can expose personal identity even if visible text is edited. PrivacyScrubber extracts document text in memory, replaces sensitive PII with tokens, strips names from the filename, and injects a clean synthetic .txt file into the AI prompt window. This eliminates 100% of binary metadata leakage while preserving full LLM analytical utility. The Free tier includes a daily allowance of 3 document uploads per 24 hours (and up to 15,000 characters per scrub), while PRO and TEAMS provide unlimited document processing.

AI Data Loss Prevention (AI DLP) refers to specialized security controls engineered to inspect and sanitize unstructured conversational prompts, streaming tokens, and file uploads before ingestion by generative AI models. Traditional gateway DLPs (like Zscaler, Netskope, or Cloudflare proxies) sit downstream in the network and struggle with encrypted WebSockets, streaming Server-Sent Events (SSE), and conversational context—often causing latency or failing to de-identify data before transit. PrivacyScrubber executes AI DLP directly at the source: inside local browser RAM (<2ms latency). Sensitive entities are converted into contextual tokens ([NAME_1], [ACCOUNT_1]) before packets ever reach the network adapter, guaranteeing zero vendor data custody and mathematical immunity to cloud leaks.

The AI Irreversibility Problem stems from the mathematical impossibility of selectively 'unlearning' or excising data once it has been digested into large language model neural weights or stored in multi-tenant RAG (Retrieval-Augmented Generation) vector databases. Machine unlearning remains an unsolved scientific challenge; post-hoc deletion requests (such as GDPR Article 17 'Right to be Forgotten') cannot be reliably enforced against fine-tuned weights or vector embeddings. Consequently, regulatory bodies (under the EU AI Act Article 10 and HIPAA Safe Harbor) require data sanitization to occur strictly prior to ingestion. PrivacyScrubber's endpoint-only RAM architecture guarantees that sensitive identifiers never enter LLM training corpora or remote vector caches.

PrivacyScrubber wraps editor inputs and tokenized outputs inside closed Shadow DOM roots. This makes unredacted sensitive text completely inaccessible to DOM-scraping scripts, third-party analytics, or unauthorized browser extensions running on the host page. Learn more about our in-DOM prompt sandbox.

Yes. The official @privacyscrubber/mcp-server package runs as a local stdio-based Model Context Protocol server for Cursor, Windsurf, Claude Desktop, and VS Code. It intercepts codebase prompts, sanitizes database passwords and API tokens in RAM, and protects your software development workflows with zero external calls.

Developers integrate @privacyscrubber/sdk with zero dependencies using a 1-line wrapper like wrapOpenAI(new OpenAI()), or via standalone sanitize() and restore() in <1ms local RAM latency. It intercepts outbound prompts before network transmission, tokenizes PII and infrastructure secrets into semantic placeholders ([NAME_1], [AWS_KEY_1]), and rehydrates LLM responses locally—including streaming SSE tokens via sliding window reconstruction. Operation is completely plug-and-play with zero background servers or Python/spaCy daemons. The primary operational risk to manage is state custody in distributed microservices: tokenMap is volatile and RAM-only, requiring stateless microservices to preserve or pass the token map between pre-processing and post-processing nodes.

No. Switching between our 30 specialized industry detection profiles (e.g. from General to Legal & Contracts or Medical) preserves active tokens and session maps. Tokens remain restorable via Reveal until you click the Clear button explicitly.

After processing documents (PDF, DOCX, XLSX), the engine immediately overwrites all temporary ArrayBuffers at the byte level with zeroes (Uint8Array.fill(0)) before releasing them to garbage collection. Pair this with our tamper-evident audit receipts for verifiable forensic compliance.

No. The platform loads zero analytics trackers, third-party cookies, or fingerprinting scripts. Maintaining strict Content Security Policies (CSP) ensures third-party endpoints cannot inspect memory or DOM states. Review our full commitments in our Zero-Server Privacy Policy.

Our Custom Regex Rules Engine discovers proprietary data formats (custom ticket IDs, project codes like PROJ-1234, internal server URLs). Token Labels, in contrast, customize the replacement placeholder (e.g. renaming [NAME] to [PATIENT_ID]) to give LLMs optimal domain context without modifying regex patterns.

PrivacyScrubber utilizes a high-performance deterministic pattern matching engine running natively in browser RAM (sub-2ms execution time). Unlike cloud LLM-based redaction which can hallucinate, omit tokens, or introduce network latency, our deterministic parser guarantees 100% reproducible tokenization across 18 core Safe Harbor/GDPR entity types (names, emails, phones, SSNs, credit cards, IBANs, medical MRNs, API tokens). PRO users can also inject custom regex rules or run local offline OCR for documents.

Because PrivacyScrubber maintains a Zero-Trust Data Sanitization (ZTDS) architecture with zero server databases and zero phone-home user tracking, license custody is sovereign to the purchaser. All license tiers are strictly non-transferable. If a key is accidentally leaked or exposed, the billing owner can immediately request emergency revocation via support@privacyscrubber.com. We append the SHA-256 hash of the compromised key to our global Key Revocation List (KRL), instantly invalidating it across all client instances within 24 hours while reissuing a secure replacement.

PRO ($15/mo or $110 Lifetime) is a single-user personal license valid on up to 3 personal devices (Web, Chrome, IDE MCP). TEAMS ($99/mo flat) is an organization-wide internal workspace covering unlimited internal staff with Master Key Isolation. The Developer SDK ($299/mo or $2,990/yr) covers headless execution across unlimited internal backend nodes, microservices, ETL pipelines, and internal RAG vector databases. Embedding into customer-facing commercial SaaS for third-party resale requires an OEM/Enterprise contract.

Yes. The @privacyscrubber/mcp-server package executes strictly in local machine memory over standard stdio JSON-RPC transport with zero outbound network requests, zero telemetry, and 100% volatile RAM mapping. It is 100% free to use with built-in session quota directly in Cursor, Windsurf, Claude Code, and VS Code without requiring a credit card or account registration. PRO and TEAMS tiers provide unlimited character volume and enterprise policy synchronization.

Solo practitioners handling confidential client sessions (ADHD coaching, therapy, executive consulting) can execute a 100% offline, zero-cloud workflow: First, transcribe your locally recorded Zoom or Teams meeting on your own device using an offline Speech-to-Text engine (such as MacWhisper or Whisper.cpp running natively on Apple Silicon / local GPU). Second, drop the resulting .txt or .docx transcript into PrivacyScrubber at privacyscrubber.com (even with your device Wi-Fi completely turned off). PrivacyScrubber strips client names, sensitive disclosures, neurodiversity/medical conditions, and identifying details into safe tokens ([NAME_1], [ORGANIZATION_1]) in volatile RAM. Third, submit the sanitized transcript to Claude or ChatGPT to extract themes, check coaching quality, or draft session summaries. Finally, click Reveal in PrivacyScrubber to rehydrate original client names locally. Zero client personal data ever touches third-party servers.

PrivacyScrubber features dedicated pattern recognition engines for European business and civil nomenclature. Our punctuation-aware lookahead models accurately tokenize initial-prefixed names with noble particles (e.g., 'J. de Ruiter', 'K. van den Berg', 'M. de Jong') without prematurely splitting or dropping initials. For statutory identity tokens, PrivacyScrubber embeds native checksum and format validators in client-side RAM for European national IDs—including the Dutch 11-test checksum for BSN/SOFI numbers, 11-digit German Steuer-ID, 16-character Italian Codice Fiscale, Spanish DNI/NIE, French NIR/INSEE, and Dutch KvK commercial register numbers.

Embedding raw personal data into vector databases (Pinecone, Qdrant, Weaviate, Milvus) creates permanent regulatory liability: fulfilling a single GDPR Article 17 'Right to Erasure' request requires locating all high-dimensional embeddings containing that person's data and recalculating vector indexes—costing upwards of $14,000 per deletion cycle. By embedding the headless @privacyscrubber/sdk into your ingestion pipeline, raw PII is swapped with deterministic cryptographic tokens in Node.js heap (<0.065ms) before vector embedding. When a deletion request arrives, organizations simply purge the local key mapping, achieving instantaneous O(1) mathematical erasure at $0 infrastructure cost.

Under GDPR Article 4(8)/Article 28 and HIPAA 45 CFR § 164.502(e), Data Processing Agreements (DPAs) and Business Associate Agreements (BAAs) are legally required only when an external vendor processes or hosts protected data on remote infrastructure on your behalf. Because PrivacyScrubber executes 100% in local client-side RAM with zero server data transmissions, zero external telemetry, and zero subprocessors, PrivacyScrubber is legally classified as a software vendor rather than a data processor. Enterprise security and compliance committees can fast-track clearance in hours using our standardized CISO Procurement Memo.

Sending text to remote Named Entity Recognition (NER) models leaks raw data before masking can occur. ZTDS executes deterministic lookaround regular expressions and 30 specialized industry vocabularies 100% locally in device memory (<1ms latency), providing 100% reproducible compliance proofs without stochastic AI hallucinations or model drift.

PrivacyScrubber enforces strict Content Security Policy (script-src 'self') with zero external scripts or analytics trackers. Session maps exist strictly in volatile, tab-isolated RAM, are never written to disk, localStorage, or cookies, and are permanently wiped on tab close. Host RAM dumping requires prior OS root compromise, which affects all workstation secrets regardless of software.

Our patent-pending engine incorporates an algorithmic Tolerant Fuzzy Unscrubbing parser that resolves model case changes, whitespace shifts (such as [NAME 1]), and markdown formatting. If a model completely omits a token from its response, the local audit view immediately flags the missing identifier without corrupting surrounding text.

Zero server processing does not mean zero auditability. PrivacyScrubber generates cryptographically verifiable CISO Audit Receipts (SHA-256 integrity hash, triggered regulatory frameworks, timestamp, entity counts) directly on the client. Enterprise teams stream zero-PII telemetry locally to internal SIEM (Splunk, Datadog) via our Node.js SDK and MCP server.

Business Associate Agreements (BAAs) provide legal recourse after a breach, not physical prevention. They do not protect against cloud logging errors, insider snooping, model jailbreaks, or US CLOUD Act subpoenas. Data that never leaves the workstation in the first place cannot be breached, inspected, or subpoenaed.

Beyond 30 pre-configured industry profiles (HIPAA, GLBA, FINRA, FAR/DFARS, GDPR), PrivacyScrubber PRO and TEAMS allow security administrators to define custom client-side regular expressions or import MDM-locked company dictionaries that execute 100% locally in device memory with zero cloud egress.