Frequently Asked Questions

Hybrid three-tier detection provides both high recall (ML-based NER for names and contextual PHI) and high precision (regex for structured identifiers). The 260+ entity types include medical-specific identifiers: MRN formats, NPI, DEA numbers, health plan IDs. Confidence thresholds can be set for maximum recall in high-risk PHI scenarios. Example: A hospital system is building a de-identified research dataset from 500,000 clinical notes. Their current tool (Presidio default) misses ~30% of PHI based on internal testing. This creates research IRB compliance issues and potential HIPAA violations. anonym.legal's hybrid approach with healthcare-specific entity types reduces the miss rate to under 5%.

Configurable confidence thresholds per entity type allow legal teams to calibrate precision vs. recall. The hybrid system's regex component provides reproducible, defensible detection for structured PII. The preview modal in the Chrome Extension shows what will be redacted before committing — the same principle applies across platforms. Example: A litigation support team at a large law firm handles 200,000-document e-discovery productions monthly. Their previous ML-only tool's 35% false positive rate exposed them to over-redaction sanctions. anonym.legal's configurable threshold system reduces false positives while maintaining privilege protection, and generates the entity-level audit log needed for privilege logs.

Confidence scoring per entity (0-100%) provides the basis for audit trails. Per-entity operator configuration allows legal teams to apply different handling rules to different entity types (e.g., replace party names with pseudonyms but redact SSNs). Reversible encryption maintains the ability to restore original text when authorized review is needed. Example: A legal technology team at a large law firm preparing document production in a commercial litigation matter. They need to redact client identifiers from 15,000 DOCX and PDF files while preserving all non-protected content. anonym.legal's hybrid detection with per-entity configuration and confidence scoring allows them to produce a defensible redaction log for the court.

Three-tier hybrid: regex handles structured data with 100% reproducibility; spaCy NLP handles contextual name/org/location detection; XLM-RoBERTa handles cross-lingual ambiguity. Confidence thresholds are configurable per entity type — a legal team can set names to 90% confidence while keeping phone numbers at regex-certainty. Example: A large law firm's e-discovery team processes 50,000 documents per litigation matter. Their ML-only redaction tool produces 35% false positive rate, requiring attorney review for each flagged item. At $400/hour and 10 false positives per document, the manual review cost exceeds the automation savings. anonym.legal's hybrid approach with configurable thresholds reduces the false positive rate to under 5%, making automation economically viable.

Confidence scoring per entity provides the audit trail foundation. The hybrid approach's use of regex for structured data makes those detections fully reproducible and explainable (exact pattern matched). NLP detections include entity type, model, and confidence — sufficient for compliance documentation. Example: A clinical research organization must demonstrate to an IRB (Institutional Review Board) that their de-identification process meets HIPAA Expert Determination standards. The audit requires documentation showing which identifiers were removed and by what method. anonym.legal's confidence scoring and entity-type classification provides the audit evidence required.

Context-aware hybrid detection with configurable thresholds per entity type. Financial-specific entity types (bank accounts, SWIFT codes, BICs, IBAN formats) use regex for deterministic detection. Names use NLP with context words and confidence scoring. Threshold configuration allows financial teams to tune for their specific volume/accuracy trade-off. Example: A digital banking platform processes 5,000 KYC applications daily across 15 European countries. Their PII detection step creates a 2-day backlog due to false positive rates requiring manual review. anonym.legal's hybrid approach reduces manual review to under 3% of documents, eliminating the bottleneck while maintaining AML compliance.

The hybrid three-tier architecture separates structured data (regex with 100% reproducibility) from contextual detection (NLP) from cross-lingual detection (transformers). Confidence thresholds are configurable per entity type. Context-aware enhancement boosts scores when context words appear near matches and suppresses false positives when context is absent. The result is dramatically lower false positive rates than Presidio defaults. Example: A data engineering team at a healthcare company running Presidio on clinical notes exported to JSON. The raw Presidio output flags hundreds of numeric sequences as SSNs and phone numbers that are actually medical record numbers, dosage amounts, and procedure codes. Manual review of false positives consumes 3+ hours per batch. anonym.legal's hybrid system with configurable thresholds and the MRN entity type reduces false positives by ~70% while maintaining PHI recall.

MCP Server intercepts all prompts sent to Claude Desktop and Cursor before they reach the AI model. API keys, connection strings, and credentials are detected (custom entity patterns support proprietary secret formats) and anonymized/redacted before transmission. The developer's workflow is unchanged — the protection is transparent. Example: A software development team at a fintech company uses Cursor IDE with Claude for code review and debugging. Their security team discovered three instances of database credentials in Claude conversation history over one quarter. Installing anonym.legal's MCP Server on developer workstations provides automatic credential scrubbing before every prompt, without requiring developers to change how they work.

MCP Server anonymizes client names, company names, deal terms, and financial figures before they reach Claude. The AI processes anonymized versions and produces output with placeholders. With reversible encryption enabled, anonym.legal automatically de-anonymizes the AI's output — the lawyer sees the original names restored in the AI response. Example: A mid-size law firm's M&A practice group uses Claude for first-pass contract review. Client names ("TechCorp acquiring MegaStartup for $450M") are replaced with tokens ("CompanyA acquiring CompanyB for $[AMOUNT]M") before Claude processes them. Claude's redlined contract comes back with the original names restored. Attorney-client privilege is preserved; AI productivity is maintained.

MCP Server acts as a transparent proxy between AI tools and the AI model. Sensitive data (source code secrets, customer PII, financial figures) is anonymized before reaching the AI. Employees continue using Claude Desktop and Cursor normally. Security teams have the control they need without productivity sacrifice. Example: A semiconductor manufacturer's security team wants to allow AI coding assistants after their competitor's Samsung-style ban hurt developer morale and productivity. They deploy anonym.legal's MCP Server on all developer workstations. Source code snippets are automatically scrubbed of credentials and proprietary algorithm identifiers before reaching Claude. AI productivity is enabled; IP protection is maintained.

Chrome Extension intercepts clipboard content before it reaches ChatGPT's input field. MCP Server intercepts at the model layer for Claude/Cursor. Both provide real-time detection with a preview modal before submission — employees see what will be anonymized and can proceed with protected data or cancel. No training required; the tool catches what employees miss. Example: A federal agency grants FOIA processing team access to ChatGPT for summarization tasks. Policy prohibits including claimant PII. The Chrome Extension intercepts any paste containing names, addresses, or SSNs and anonymizes them before they appear in the ChatGPT input field. Contractors can use AI for efficiency without accidental PII exposure.

Chrome Extension installs in minutes and immediately intercepts PII before it reaches ChatGPT, Claude.ai, and Gemini. No DLP configuration required. MCP Server for Claude Desktop and Cursor requires minimal setup. Both tools work without network-level changes, making them deployable on individual workstations or enterprise-wide via policy. Example: A 200-person professional services firm learns from industry news that 83% of organizations lack AI controls. Their CISO wants to implement controls within 30 days without a major IT project. anonym.legal Chrome Extension is deployed to all workstations via Chrome Enterprise policy in one afternoon. The MCP Server is installed for the development team. Full AI PII protection deployed in hours, not months.

The MCP Server on port 3100 acts as a transparent proxy. All text passed to Claude Desktop or Cursor through the MCP protocol is filtered for PII before reaching the AI model. Developers configure once; protection is automatic. All 5 anonymization methods are available — developers can use reversible encryption to pseudonymize code identifiers (e.g., customer IDs in database queries) and decrypt AI responses automatically. Example: A senior developer at a healthcare SaaS company using Cursor to write database migration scripts. The scripts contain patient record IDs, database connection strings, and proprietary data models. The MCP Server intercepts the prompt, replaces sensitive identifiers with encrypted tokens (using reversible encryption), and sends the clean prompt to Claude. The AI response arrives with tokens; the MCP Server auto-decrypts to restore original context. Developer productivity is preserved; PHI never reaches Anthropic's servers.

The MCP Server provides exactly this technical control layer. It sits between the user's AI tool and the AI model API. All prompts pass through the anonymization engine; sensitive data is replaced/encrypted before transmission. Security teams get audit trails. Developers get AI productivity. The reversible encryption option means responses from the AI can reference the pseudonymized data and be automatically decrypted for the developer's view. Example: The CISO at a German automotive manufacturer needs to enable AI coding assistance for 500 developers while complying with GDPR and protecting trade secrets (proprietary manufacturing algorithms in the codebase). The MCP Server deployment filters all prompts through anonym.legal's engine before they reach Claude/Cursor APIs. Security team approves; developers keep AI access; IP stays protected.

AES-256-GCM reversible encryption preserves the mathematical relationship between the anonymized token and the original value. With the client-held encryption key, any anonymized document can be fully restored to its original content. Without the key, the anonymized version is computationally indistinguishable from a permanently redacted document. Legal teams share encrypted versions; produce originals when required using the retained key. Example: A pharmaceutical company shares clinical trial data with external statisticians using anonym.legal's encrypted anonymization. Two years later, the FDA requests original patient records as part of a drug safety review. The company restores the original data using their retained encryption key — no spoliation, no missing records, full regulatory compliance. The statisticians' encrypted copies remain protected throughout.

Reversible encryption creates a protected pseudonymization layer. The research dataset uses encrypted tokens. The decryption key is held by the designated data custodian. When re-contact is clinically justified and IRB-approved, the custodian decrypts the specific participant records to enable follow-up. The broader dataset remains protected — only the specific authorized decryption is performed. Example: A European oncology research center conducts a 5,000-patient study using anonym.legal's encrypted anonymization. Mid-study analysis reveals a subgroup of 47 participants showing markers for an aggressive cancer variant. The ethics committee approves re-contact. The data custodian uses the retained encryption key to identify the 47 real patients. Those patients are contacted, 23 are found to have actionable findings. The remaining 4,953 participants' data remains fully protected.

Reversible encryption using AES-256-GCM generates deterministic encrypted tokens from original PII. The key is held only by the user. "John Smith" becomes "[ENC:x9f3a...]" consistently throughout the document — maintaining referential integrity. When authorized de-anonymization is needed (discovery production, audit verification, research follow-up), the user applies their key and all tokens restore to originals. The Chrome Extension auto-decrypts AI responses, so working with encrypted data is transparent in the AI workflow. Example: A compliance officer at a pharmaceutical company shares clinical trial data with a contract research organization (CRO). All patient identifiers are encrypted with a company-held key. The CRO analyzes anonymized data. When the FDA requests original patient records for audit, the compliance officer applies the key and produces the originals in minutes — with a cryptographic audit trail proving chain of custody.

Reversible encryption allows selective de-anonymization. The finance team shares encrypted anonymized reports. Auditors working under formal engagement can be given decryption capability for their audit period. After audit completion, the key can be rotated — previous encrypted copies remain protected, auditors cannot retroactively access records outside their engagement. Example: A private equity firm shares portfolio company financial data with an external audit firm for annual review. Client company names and deal terms are encrypted before sharing. During audit, the engagement partner receives temporary decryption access for the audit period. After the audit opinion is issued, key rotation removes that access. Former employees of the audit firm cannot access the data after their tenure.

Reversible encryption allows HR to run "conditionally anonymous" surveys. Responses are encrypted before storage. The decryption key is held by a designated HR executive (or third-party ombudsman). When a response contains a serious allegation meeting predefined criteria (e.g., physical harassment, legal violations), the authorized party can decrypt that specific response to identify the reporter and initiate formal investigation. Example: A 2,000-employee manufacturing company's annual culture survey captures an allegation of serious misconduct by a senior executive. The response is encrypted. The company's third-party ombudsman reviews the allegation and determines it meets the threshold for de-anonymization under the company's published survey policy. The ombudsman decrypts the specific response, contacts the reporter through a formal protected channel, and initiates an independent investigation. All other responses remain permanently anonymized.

Session-based token mapping maintains consistent anonymization within a conversation. The same customer name always maps to the same token within a session. Auto-decrypt in Chrome Extension responses restores real names in AI outputs before display. Persistent token mapping is also available for longer-lived workflows. Example: A German insurance company's AI-powered claims processing system processes customer complaint emails. Customer names, policy numbers, and claim amounts are anonymized before Claude processes the emails. Claude drafts a response using the anonymized tokens. anonym.legal's auto-decrypt restores original customer information in Claude's draft before it is displayed to the claims handler. The handler sends the final response with real customer names. GDPR compliance is maintained throughout.

Reversible encryption generates consistent tokens (deterministic AES-256-GCM) — "Patient_001" maps to the same encrypted token throughout all study records. The research team holds the key. Re-identification for follow-up requires the key holder to decrypt. All decrypt events are logged. This satisfies both the IRB requirement for controlled re-identification capability and the HIPAA Safe Harbor requirement for de-identified data sharing.

Custom entity creation with AI-assisted regex generation is purpose-built for this use case. A compliance officer describes the MRN format ("Hospital identifier starting with HOSP, dash, 4-digit year, dash, 6-digit number") and receives a working regex pattern. Custom entity is saved, applied to all document processing, and shared with the team via presets. Zero engineering required. HIPAA Safe Harbor compliance for organization-specific identifiers is achievable in under an hour. Example: A regional hospital network (15 facilities) is preparing to share de-identified patient data with a university research partner. Their MRN format (HOSP-YYYY-XXXXXX) appears in thousands of discharge summary PDFs. Their compliance team uses anonym.legal to define the custom MRN pattern, validate it against a sample document set, and process the full research dataset in batch. The university receives HIPAA-compliant de-identified data. Compliance timeline: 3 days vs. 3 months for custom code developmen

Custom entity creation with AI-assisted pattern generation. Users describe their identifier format in plain language ("Employee IDs that start with EMP followed by 5 digits") and the AI generates the appropriate regex pattern. Custom entities integrate seamlessly with the existing 260+ type detection. Results can be saved as presets and shared across teams. Zero engineering required — compliance and legal teams can define their own patterns. Example: A financial services firm has customer account numbers in the format "ACC-XXXXXXXX-XX" that appear throughout support ticket exports. Standard PII tools miss them entirely. Using anonym.legal's custom entity builder, their compliance team creates a pattern in 10 minutes. All 180,000 historical support tickets processed in batch now have account numbers redacted alongside standard PII. Re-identification risk eliminated without an engineering ticket.

The 260+ entity library includes major European national identifiers. For formats not yet covered, the custom entity builder allows compliance teams to add them using the AI pattern assistant or manually entering the regex. Once added, they're available in all processing modes and can be shared via presets to the entire team. The German Steueridentifikationsnummer, for example, can be added in under 5 minutes. Example: A German payroll outsourcing firm processes documents for 500 client companies. Their anonymization workflow missed Steueridentifikationsnummern in payslip PDFs because their previous tool (standard Presidio) had no German tax ID recognizer. After a DPA audit finding, they need to add this detection immediately. anonym.legal's custom entity creation lets their compliance officer add the pattern without waiting for an engineering sprint — critical gap closed in one afternoon.

Custom entity creation for order IDs and account numbers in specific formats, combined with the default 260+ entity type detection, provides complete anonymization in a single pass. The Chrome Extension or MCP Server can apply custom entity detection in real-time as support agents type — preventing PII and custom identifiers from ever reaching external AI systems. Configuration is shareable across the support team via presets. Example: A SaaS company's customer support team uses Claude via their internal AI platform to draft support responses. Customer messages copied into the AI interface contained customer names, email addresses, and order IDs (ORD-XXXXXXX format). After a GDPR review, the DPO required anonymization before AI processing. anonym.legal's Chrome Extension with custom order ID entity detects and replaces all identifiers in real-time. Support team workflow unchanged, GDPR compliance achieved.

Custom entity creation supports legal identifier formats. Attorneys and compliance officers can define bar number formats (State + 6 digits), docket number formats (XX-CV-XXXXXX for federal civil), and matter number formats using the AI-assisted pattern builder. These custom entities integrate with standard PII detection, enabling comprehensive document review. The resulting preset can be shared across the legal team or sold as a product feature by legal tech vendors integrating via API. Example: A legal AI startup builds a document analysis tool for law firms. Their enterprise clients require redaction of client matter numbers alongside standard PII before documents are processed by their AI. Using anonym.legal's custom entity API, they add matter number detection to their pipeline in 2 days (vs. 3 months building a custom NLP model). Their enterprise contracts close without the compliance blocker.

The AI-assisted pattern helper accepts plain-language examples ("These look like MRN numbers: MRN:1234567, MRN:9876543") and generates the appropriate regex pattern. The visual regex builder allows refinement. The test interface validates against sample text. Patterns are saved as named custom entities and can be shared across the team with Basic+ plans.

The hybrid recognizer stack (Regex + NLP + XLM-RoBERTa transformers) dramatically improves precision by using context from surrounding text. Transformer-based models understand that "Apple announced its earnings" refers to a company, while "Apple Smith joined the team" refers to a person. The result is materially higher precision than bare Presidio, preserving document utility while maintaining privacy protection. Users who experienced Presidio's false positive problem find anonym.legal's accuracy meaningfully better. Example: A data analytics firm processing customer feedback surveys abandoned Presidio after 40% of survey responses had product names, city names, and brand mentions incorrectly redacted alongside actual PII. Downstream analysis was corrupted by over-anonymization. Switching to anonym.legal's hybrid recognizer, precision improved to ~85%+ — product names preserved, person names correctly identified. Analysis quality restored.

anonym.legal is the managed version of the Presidio engine with significant extensions. Zero setup, zero infrastructure, zero maintenance. Users get Presidio's NLP accuracy (plus XLM-RoBERTa improvements) through a web interface, desktop app, or API — without touching Docker, Python, or spaCy model downloads. The Desktop app provides offline capability for air-gapped environments without the complexity of self-hosted Presidio. Example: A compliance team at an insurance company spent 3 days trying to get Presidio running in their environment. After a Docker networking issue caused the 4th crash, the project was escalated. anonym.legal was evaluated as an alternative: sign-up to first anonymization run in 12 minutes. The insurance company adopted anonym.legal Professional at €180/year. Estimated engineering time saved vs. managing self-hosted Presidio: 60 hours initial setup + 72 hours/year maintenance = ~132 hours of engineering time at €100/hour = €13,200 saved vs. €180 cost.

260+ entity types built on the Presidio foundation include comprehensive European identifier coverage: IBAN numbers, European driving license formats, EU member state tax identifiers, national health numbers, social insurance numbers, and VAT numbers for major EU economies. This coverage is maintained, tested, and updated as regulations and formats change — without requiring open-source contribution effort from users. Example: A German fintech handling EU customer financial data needs to detect IBANs, BICs, German tax IDs, and German commercial registration numbers (Handelsregisternummer) in customer documents. Presidio detects 0 of these 4 entity types out of the box. Writing and maintaining custom recognizers for all 4 requires 20-40 engineering hours plus ongoing testing. anonym.legal includes all 4 plus 256 additional entity types at €180/year.

The managed SaaS model eliminates all production deployment concerns — scaling, monitoring, failure handling, and audit logging are handled by anonym.legal's infrastructure. Users get SLA-backed availability, automatic scaling, and comprehensive audit trails without building any of this infrastructure themselves. The Desktop app provides offline processing for air-gapped environments without requiring production server management. Example: A healthcare SaaS company's engineering team spent 6 weeks attempting to build a production-grade Presidio deployment for their PHI anonymization pipeline. After repeated failures with model loading timeouts and inconsistent API behavior under load, the team evaluated managed alternatives. anonym.legal's API endpoint replaced the self-hosted deployment in 3 days. Engineering time reclaimed: 6 weeks × 2 engineers = 12 engineering weeks ($48,000+ at US rates). Annual anonym.legal Business plan: €348.

anonym.legal provides Presidio's detection capabilities (extended to 267 entities and 48 languages) as a fully managed service with no infrastructure management required. The web, desktop, Office, Chrome, and MCP interfaces make the underlying Presidio engine accessible to non-technical users. Continuous updates maintain accuracy without requiring teams to manage model versions. The free tier allows evaluation without commitment.

As a managed SaaS and Desktop product, anonym.legal maintains consistent model versions across all user environments. There's no staging vs. production discrepancy — all users run the same engine version at the same time. Desktop app users get the same engine as web users. Updates are managed centrally and versioned explicitly. Compliance auditors see consistent, reproducible behavior documentation rather than environment-specific variability. Example: A financial services firm's data engineering team discovered their Presidio staging environment (spaCy 3.4.4) was producing different NER results than production (spaCy 3.5.1). An audit found 3% of documents were differently anonymized in production vs. their test results. Migrating to anonym.legal eliminated environment-specific variation — the same managed engine runs everywhere. Audit finding closed.

The Chrome Extension provides real-time PII detection with inline highlighting directly in the ChatGPT, Claude, and Gemini input fields. Detection happens client-side before data is submitted. Highlighted PII can be anonymized with one click before submission. The user sees which entities were detected and their confidence scores, enabling informed decisions about what to share. Prevention at the point of entry, not detection after the fact. Example: A law firm's associates use Claude to draft contract summaries. The Chrome Extension highlights client names, case numbers, and financial figures in the Claude input field before submission. Associates can anonymize with one click before sending. In 6 months of deployment, zero client PII incidents vs. 3 incidents in the previous 6 months (before extension deployment). The managing partner credits the real-time prevention model for the improvement.

The Chrome Extension provides per-user, per-session detection metrics that feed into organizational visibility dashboards. IT administrators can see anonymization activity across deployed users: total PII entities detected, entity types, AI platforms used, and anonymization rate (how often detected PII was anonymized before submission vs. ignored). This provides the monitoring data compliance teams need to demonstrate appropriate measures under GDPR Article 32. Example: A financial services firm's CISO needs to demonstrate to auditors that AI tool PII exposure is monitored and controlled. anonym.legal Chrome Extension deployed to 500 employees generates organizational dashboards showing: 12,000 PII detections per week, 94% anonymization rate, top entity types (customer names, account numbers, transaction IDs), and the 6% of detections submitted without anonymization (flagged for follow-up training). Auditors receive quantitative evidence of active monitoring and control.

Confidence scoring per entity (0-100%) allows configurable thresholds. Entity highlighting in the source text provides visual feedback before any action is taken. The Chrome Extension's pre-submission interception is architecturally prevention-first: the prompt never reaches the AI model unless the user explicitly proceeds. Real-time detection in the web/desktop UI provides instant feedback as text is entered.

Real-time detection with confidence scoring operates on any text input. The 260+ entity types include all 18 HIPAA PHI identifiers. Detection can be integrated at the clinical documentation review stage before EHR commit. The preview modal shows detected entities, allowing clinical staff to review before proceeding.

Every detected entity displays a confidence score with visual indicators (high/medium/low). Users can set confidence thresholds: entities above 85% confidence are auto-anonymized; entities between 50-85% are flagged for human review; entities below 50% are surfaced as suggestions. This creates an auditable, defensible anonymization workflow that satisfies compliance documentation requirements and reduces both false positives (over-redaction) and false negatives (missed PII). Example: A legal discovery firm processes client documents where over-redaction is as problematic as under-redaction — redacting attorney names or court references corrupts the legal record. Using anonym.legal's confidence threshold settings (auto-redact above 90%, review 60-90%, ignore below 60%), they create an auditable workflow where attorneys review only medium-confidence detections. Review time drops by 65% vs. manual review of all detections, while the audit trail documents exactly which entities were auto-r

Real-time detection capabilities (via Chrome Extension inline detection or MCP Server API integration) can be integrated into web applications to validate form inputs before submission. The Chrome Extension works on any web form in the browser. For custom application integration, the MCP Server API provides real-time PII detection that can be called on form submit events. Both provide confidence scores for entity-level decision making. Example: A healthcare patient portal allows patients to submit "free text" symptoms descriptions. The form regularly receives entries containing other patients' names (caregiver descriptions) and social security numbers (insurance reference). Integrating anonym.legal's real-time detection via the API, the portal now warns patients before submission if their input contains PII in unexpected fields. GDPR data minimization compliance improved; database PII contamination reduced by 80%.

The Chrome Extension activates automatically on paste events in supported AI interfaces (ChatGPT, Claude, Gemini). When a user pastes text containing PII, entities are highlighted immediately without any user action. A one-click anonymization button replaces highlighted entities. The user's workflow: paste, notice highlights, click anonymize, submit. The "remember to check" step is eliminated — the visual highlight is the reminder. Example: A customer success team of 30 agents at a B2B SaaS company uses Claude to summarize customer call notes. Before the Chrome Extension deployment, the team lead estimated 15-20 PII incidents per month (customer names and company details in Claude prompts). After 90-day deployment of anonym.legal Chrome Extension, reported incidents dropped to 1-2 per month. The team lead attributes the improvement to "the highlights make it impossible to ignore."