Layrin
Pricing

AI PRIVACY GUIDE

Protect sensitive text before it reaches AI.

People often look for ways to anonymize text before ChatGPT, redact PII from an AI prompt, or hide confidential information before using an AI assistant. Layrin replaces sensitive values locally with structured placeholders, keeps the protected text reviewable, and restores the original values on your device after the AI responds.

The same workflow works with a short email, a detailed report, meeting notes, a contract excerpt, clinical or research text, HR information, and other professional content pasted into Layrin.

EnglishFrançais

Table of contents

Why protect sensitive text before using AI?What is LLM data masking?What anonymization, redaction, masking, and tokenization meanHow Layrin’s local-first workflow worksWhere sensitive originals remainPrivacy tokens and AI model tokensWhat types of sensitive information can be protected?From short emails to longer professional reportsLayrin Protect and Layrin ProWhy human review still mattersCurrent scope and responsible useFrequently asked questions

1. Why protect sensitive text before using AI?

AI tools can help draft, summarize, translate, organize, and analyze professional text. The same text may also contain names, contact details, dates, financial values, client information, internal identifiers, credentials, project references, or other confidential details that are not necessary for the AI task.

A privacy step before AI use helps separate the useful meaning of the text from the sensitive values inside it. Instead of sending every original value, you can work with a protected version that preserves context through placeholders such as [PERSON_1], [DATE_1], [AMOUNT_1], or [SECRET_1].

This is useful when the AI needs to understand the structure and purpose of the text but does not need the real identity, account number, access credential, or confidential reference.

2. What is LLM data masking?

LLM data masking is a broad search term for replacing sensitive information before unstructured text reaches a large language model. It can include PII masking for AI, hiding confidential business values, or substituting data that ChatGPT or another AI service does not need to see.

Traditional data masking often operates on structured database fields or test datasets. Layrin is instead a local-first privacy gateway for unstructured professional text: it uses reversible local tokenization and contextual placeholders, keeps the mapping on the device, and restores supported values locally after the AI responds.

This preserves useful relationships in a prompt without describing Layrin as a database-masking platform or claiming that the resulting text is irreversibly anonymous.

For a closer comparison of these techniques, read Data Masking vs Tokenization.

3. What anonymization, redaction, masking, and tokenization mean

The terms anonymization, redaction, masking, and pseudonymization are often used interchangeably in searches, but they do not describe exactly the same process.

Irreversible anonymization

True anonymization generally aims to prevent information from being linked back to an identifiable person. It is not designed for restoring the original values later.

Redaction

Redaction removes or obscures information. A permanently redacted value may no longer be available in the resulting text.

Masking

Masking hides all or part of a value, such as displaying only the final digits of an account number.

Reversible tokenization

Layrin uses reversible local tokenization. Sensitive values become structured placeholders, while original values and their mappings remain on your device for local restoration.

Unlike irreversible anonymization, this workflow intentionally preserves a local mapping so supported values can be restored after the AI responds.

Comparison of data masking, redaction, anonymization, pseudonymization, and reversible tokenization
TechniqueOriginal valueRestorationTypical purposeRelationship to Layrin
Data maskingHidden, altered, or substituted; implementations varySometimes, depending on the method and retained dataLimit exposure in displays, datasets, or workflowsRelated term, but not a complete description of Layrin
Data redactionRemoved or obscured in the resulting contentUsually not from the redacted copy alonePrevent disclosure in a released documentLayrin does not permanently redact the working copy
Irreversible anonymizationTransformed or removed to prevent re-identificationNoCreate data that is no longer linked to an identifiable personNot what Layrin claims to perform
PseudonymizationReplaced with a pseudonym or referencePossible with separately held informationReduce direct identifiability while retaining utilityConceptually related to controlled placeholder substitution
Reversible tokenizationReplaced with a token linked through a controlled mappingYes, for an authorized workflow with the mappingPreserve context while withholding the original valueLayrin uses reversible local tokenization and contextual placeholders

To decide when information should be transformed or removed, compare data masking and data redaction.

For the broader family of techniques used across databases, logs, APIs, documents, and prompts, read the guide to PII masking.

To understand the identifiability and reversibility boundary, compare data masking with anonymization, then review the broader guide to data anonymization techniques.

4. How Layrin’s local-first workflow works

The workflow follows four clear stages:

Protect locally

Paste the text into Layrin. The local protection engine identifies sensitive values and replaces them with numbered placeholders. Original values and their mappings remain on the device.

Review

Review the protected text before it is used with AI. You can verify the detected values and manually protect text that requires additional attention. Review is important because professional context can be subtle and no automated detector should be treated as infallible.

Use AI

With Layrin Protect, copy the reviewed protected text into the AI tool or workflow you already use. With Layrin Pro, keep the protected drafting workflow inside Layrin through the managed AI experience.

Restore locally

After the AI response is available, Layrin restores recognized placeholders with the original values on the device. The restored result can then be reviewed and copied for its intended use.

Stage 1

Original professional text

Sarah Martin from Meridian Health Analytics asked us to send the agreement to sarah@example.com.

Stage 2

Locally protected text

[PERSON_1] from [COMPANY_1] asked us to send the agreement to [EMAIL_1].

Stage 3

AI output using protected tokens

Draft a reply confirming that [PERSON_1] from [COMPANY_1] will receive the agreement at [EMAIL_1].

Stage 4

Locally restored response

Draft a reply confirming that Sarah Martin from Meridian Health Analytics will receive the agreement at sarah@example.com.

Protect locally → Review → Use AI → Restore locally

5. Where sensitive originals remain

Layrin’s local-first boundary is designed around the original sensitive values:

  • Sensitive originals and token mappings remain on the device.
  • Sensitive values are replaced locally with contextual placeholders.
  • The user reviews the protected text before choosing to use it with AI.
  • Only reviewed protected text may be sent to the selected AI service.
  • Supported placeholders in the AI response are restored locally.

This boundary reduces unnecessary disclosure, but it does not guarantee perfect detection, anonymity, compliance, or zero leakage. The protected text and the selected AI workflow still require review and judgment.

Private AI can mean running the model itself locally or privately. Layrin takes a different approach: it keeps sensitive values and restoration mappings local while allowing reviewed protected text to be used with the AI service selected by the user. It is a local-first, privacy-first protection layer for professional text—not a private or self-hosted LLM, an encrypted AI chat, an on-device AI assistant, or a private ChatGPT replacement.

For the broader ChatGPT privacy, shadow AI, and organizational risk picture, see how AI data leakage happens and how different controls address it.

For ChatGPT-specific training, retention, conversation access, managed-account, and privacy-control details, read the guide to ChatGPT privacy.

6. Privacy tokenization is not AI model tokenization

Privacy or security tokenization replaces a sensitive value with a controlled token such as [PERSON_1] or [ACCOUNT_1]. Its purpose is to withhold the original value while preserving enough context for an authorized workflow and, when appropriate, local restoration.

NLP or model tokenization is different. An LLM internally splits text into model tokens for processing, context windows, and usage calculation. Those model tokens do not remove sensitive information and should not be confused with a privacy control for AI prompts.

Layrin’s placeholders protect selected values; an LLM’s model tokens are units used to process text.

For the broader definition, architecture, and non-AI use cases, read the guide to data tokenization.

7. What types of sensitive information can be protected?

Depending on the wording and context, Layrin is designed to detect or help you manually protect information such as:

  • People and contact information, including names, email addresses, and phone numbers
  • Dates, locations, and postal addresses
  • Financial values, account details, and business identifiers
  • Legal, clinical, research, and project references, including matter, participant, and specimen identifiers
  • Credentials, secrets, access keys, URLs, paths, hosts, and other technical resources
  • Other confidential values selected during review

Detection results depend on the text, language, formatting, and context. Layrin’s review and manual-protection controls are part of the privacy workflow, not optional decoration.

8. From short emails to longer professional reports

Layrin is not limited to a single sentence or a short email. The same protect-and-restore workflow can be applied to longer professional text pasted into the application.

Emails and messages

Protect names, contact details, dates, amounts, confidential references, and source signatures before drafting or revising a message.

Reports and meeting notes

Replace participant names, projects, decisions, budgets, timelines, and internal references while preserving the structure needed for summarization or analysis.

Contracts and legal drafts

Protect client information, parties, matter references, contract identifiers, financial values, addresses, and selected confidential clauses in pasted legal text or contract excerpts.

Clinical and research text

Protect patient-related references, participant identifiers, dates, study information, specimen references, and research project details before using AI for an appropriate reviewed task.

HR and employee information

Protect candidate names, employee details, compensation values, workplace references, dates, and internal identifiers in recruiting or workplace text.

Financial, operational, and technical text

Protect amounts, account details, business identifiers, project codes, infrastructure references, secrets, and operational information before requesting AI assistance.

Layrin currently works with text pasted into the application. Direct file import and OCR are not currently offered, including PDF, Word, spreadsheet, and image import.

9. Layrin Protect and Layrin Pro

Layrin Protect

Protect is for people who want to keep using their existing AI tools. Layrin protects and reviews the text locally, you copy the protected version to the AI workflow you choose, and you restore the response locally in Layrin.

Layrin Pro

Pro provides the integrated managed AI workflow. The text is protected and reviewed before the managed request, and original sensitive values and local mappings remain on the device. The protected content required for the task is processed through the managed Layrin service, then the response is restored locally.

Both plans use the same core principle: original sensitive values should not be included in the AI task when a protected placeholder can preserve the necessary context.

10. Why human review still matters

Professional context can make a project nickname, unusual identifier, internal reference, or organization-specific phrase confidential even when no clear label announces it.

Review the protected text before AI use and correct any missed value, ambiguous token, or classification that needs additional judgment.

11. Current scope and responsible use

Layrin currently works with pasted text as a local-first privacy gateway. It is not a general chat application or a substitute for professional legal, clinical, security, or privacy review.

  • Detection can miss or overprotect a value.
  • Use manual protection for values requiring additional judgment.
  • Compliance with laws, policies, contracts, and professional obligations still depends on the user, organization, task, and chosen workflow.
  • Choose an AI provider and workflow appropriate for your organization and task.

Use professional judgment for the content, AI output, and decisions involved in the task.

12. Frequently asked questions

What is LLM data masking?

LLM data masking is a broad phrase for replacing sensitive information before text reaches an AI system. Layrin specifically uses reversible local tokenization: contextual placeholders replace sensitive values, mappings remain on the device, and supported values can be restored locally.

Should I mask PII before using ChatGPT?

When real identities are not needed for the task, replacing personal and confidential values before using ChatGPT can reduce unnecessary disclosure. Review the protected text first and use a service and workflow appropriate for your organization and task.

Is tokenizing a prompt the same as AI tokenization?

No. Privacy tokenization substitutes sensitive values with controlled placeholders. AI or NLP tokenization splits text into the model tokens an LLM processes for context and billing; it is not a privacy control.

Can protected AI responses be restored?

Yes. When an AI response preserves supported Layrin placeholders, Layrin can replace them with the corresponding original values locally on the device.

What kinds of confidential text can Layrin protect?

Layrin is designed for pasted professional text such as emails, reports, meeting notes, contract excerpts, clinical or research text, HR information, and operational content. It can protect many personal, financial, legal, technical, and organization-specific values.

Is sensitive-data detection guaranteed to find everything?

No. Detection is not guaranteed to identify every sensitive item. Review the protected text before using it with an AI service, and manually protect any additional value that requires attention.

Add a privacy step before your next AI task.

Protect sensitive text locally, review the result, use AI with clearer boundaries, and restore original values on your device.

Compare plansStart your 7-day trial

© 2026 Priova Intelligence. All rights reserved.

Layrin is a privacy product by Priova Intelligence.

PrivacyTermsSecurityGuidesResearchContact