Confidential AI chat · Prompt anonymization

Use AI without submitting sensitive data to LLMs.

Hebed Chat anonymizes client names, addresses, IBANs, amounts and matter details before your prompt reaches ChatGPT, Claude, Gemini or Copilot. Same chat experience you already use — with less exposure and a record of what almost got sent.

Test a prompt→See how it works
Supported & developed with
·HEG Genève·HES·SO·Swiss Confederation

You use AI tools every day for drafting, research and review. No one is checking behind you.

Independent lawyers and consultants use ChatGPT, Copilot and Harvey for drafting and research every day — an environment where prompts often include information that would not be appropriate to share externally. Hebed Chat helps organizations maintain better control over what leaves their environment.

How it works

One extra step, invisible to your workflow.

You keep typing prompts the way you always have. Hebed does the anonymization in between — you never have to write a "safe" version by hand.

01

Write your prompt

Use Hebed Chat exactly like ChatGPT — drafting, summarizing, rewording, translating.

02

Entities are detected

Client names, addresses, IBANs, amounts and matter details are flagged in real time, before anything is sent.

03

Prompt is anonymized

Sensitive elements are replaced with safe placeholders. Only the anonymized version reaches the model.

04

Risk is logged

Every detection is categorized, giving you visibility into what your team almost exposed — and how often.

Example — private client / wealth matter
"Reword this note for Jean Martin regarding the restructuring of his holding, assets of CHF 12.4M, and beneficiaries Marie and Luc Martin."
Flagged. Three personal names and one financial amount detected.
Example — corporate / M&A
"Prepare a summary of the term sheet between Acme SA and Beta Holding, purchase price CHF 3.2M, earn-out CHF 750k, CEO Pierre Blanc."
Flagged. Two company names, two amounts and one personal name detected.

Sample prompts above are synthetic and for demonstration only.

What gets anonymized

Built to catch the details that matter most in legal work.

Hebed's detection engine is built around entity-recognition and anonymization techniques adapted for legal and professional-services content, not generic web text.

Client & personal names
Physical addresses
IBANs & bank details
Financial amounts
Company & entity names
Matter & case-specific details

Built on regulation, not a guess

The detection rules are built around the personal-data categories defined in PII regulation such as GDPR and the Swiss nLPD, then extended with legal-specific patterns for client names, matter references and case facts — rather than a generic keyword list.

Detection is not infallible. Hebed lets you review flagged entities before sending, and legal-specific rules are refined based on real usage. We do not claim perfect anonymization for every prompt.

Straight answer

What Hebed Chat does today and what it doesn't, yet.

We'd rather you know exactly what you're getting. Hebed Chat is a confidential chat interface, not a full enforcement layer across every device and tool you use. That's coming, but it isn't live yet.

Live today

  • →Anonymizes sensitive prompt content before it reaches the model
  • →Detects and categorizes risk signals for visibility
  • →Familiar chat interface for drafting, summarizing, rewording, translating
  • →Works across common models: ChatGPT, Claude, Gemini, Mistral, Copilot
Common question

"Why not just use ChatGPT Enterprise or Copilot?"

Enterprise AI plans improve how a provider stores and trains on your data. They don't remove the client name, IBAN or matter detail you typed into the prompt itself.

Capability
ChatGPT Enterprise / Copilot
Hebed Chat
Improves data storage & training terms
Yes
Not applicable — model-agnostic
Removes client names, IBANs, addresses from the prompt itself
No, by default
Yes, before transmission
Categorizes risk for visibility/reporting
Limited
Yes
Works the same across multiple models
Single provider
Yes — ChatGPT, Claude, Gemini, Mistral, Copilot
Who it's for

Built for independent practitioners who can't just paste and send.

No compliance department, no IT team behind you — every role below still handles privileged client information and wants to use AI for everyday drafting, summarizing and translation without adding exposure.

Solo practitioner

Independent lawyer

Draft and reword client correspondence with AI without a firm's compliance layer behind you.

"Can I use AI safely without exposing client facts, on my own?"
Legal consultant

Independent counsel / of counsel

Show clients a concrete, demonstrable example of prompt-level anonymization in your own advisory work.

"Can I demonstrate this safeguard to a client directly?"
Data protection consultant

Freelance privacy & DPO advisor

Advise regulated clients on AI use while keeping your own drafting workflow exposure-free.

"How do I advise on this risk if I can't demonstrate a fix myself?"
Tax & private client advisor

Independent tax practitioner

Draft and summarize client communications involving IBANs, holdings and beneficiaries without leaking them.

"Can I use AI on wealth files without exposing the client?"
Corporate & M&A consultant

Independent deal advisor

Summarize term sheets, valuations and due-diligence notes without exposing deal parties or figures.

"Can I run deal documents through AI before signing is public?"
Employment & family law consultant

Independent practitioner

Handle highly personal client facts in drafting without them ever reaching a public model.

"This file is sensitive — can AI actually help me here?"
Works with

The models you already use.

Hebed Chat anonymizes the prompt, then sends the safer version to the model of your choice.

Claude
ChatGPT
Gemini
Copilot
Mistral
HHarvey
Spellbook
LLegora

This list reflects the models Hebed Chat is designed to route anonymized prompts to. It does not imply a partnership or endorsement by these providers.

From chat to governance

Anonymization is the first step, not the destination.

Hebed Chat is Phase 1. Each phase after it builds toward a full AI governance layer for your practice — visibility first, enforcement later.

1
Anonymize & Observe

Detect and anonymize sensitive prompt content; capture risk categories.

You are here
2
Warn & Educate

User-facing warnings and safer prompt suggestions before sending.

3
Sharing & Controls

Optional sharing with co-counsel, basic reporting, simple access controls.

4
Enforce

Blocking, reverse proxy/gateway, integrations across every tool you use.

5
Govern

Audit trail, compliance mapping, full AI governance infrastructure.

FAQ

Questions lawyers ask before trying it.

What is prompt anonymization?
Prompt anonymization means detecting and masking sensitive elements — client names, addresses, IBANs, amounts, matter details — inside a piece of text before it's sent to an AI model. The model answers the question without ever seeing who or what the prompt was really about.
Is Hebed Chat the same as ChatGPT Enterprise or Microsoft Copilot?
No. Enterprise plans govern how a provider stores and trains on your data. They don't remove a client's name or IBAN from the text you actually typed. Hebed Chat sits in front of the model and anonymizes the prompt itself, regardless of which provider answers it.
Which sensitive data can Hebed Chat detect?
Client and personal names, physical addresses, IBANs and bank details, financial amounts, company names, and matter-specific confidential details, using entity-detection techniques adapted for legal content.
Is the anonymization engine trained specifically for legal data?
The detection rules are built around the personal-data categories defined in PII regulation such as GDPR and the Swiss nLPD, then extended with legal-specific patterns for client names, matter references and case facts.
Where is my data stored?
Prompts and anonymization logs are stored with a Swiss cloud provider, consistent with Swiss data residency and nLPD expectations.
Is the anonymization perfect?
No detection system catches everything. Hebed Chat flags likely sensitive elements and lets you review before sending. Rules are refined for legal use cases over time, but we don't claim perfect anonymization for every prompt, and highly sensitive matters should still be reviewed.
Which AI models does Hebed Chat work with?
It's built to route anonymized prompts to commonly used models, including ChatGPT, Claude, Gemini, Mistral and Copilot, so you keep a familiar chat experience.
Is Hebed Chat a firewall or a full AI governance platform?
Not yet. Today it's a confidential chat interface with anonymization and risk visibility — it doesn't block every AI tool you use or intercept traffic through a reverse proxy. Broader enforcement controls are on the roadmap, not live today.
How do I get started?
Test a prompt for free at hebedai.com/chat — no setup required to see how anonymization works. If you want to trial it in your own practice, ask about a 30-day pilot.
Try it now

Test a prompt. See what gets caught.

Paste a realistic legal prompt and watch Hebed Chat detect and anonymize the sensitive parts before anything reaches a model. No setup, no commitment.

Test a prompt free →Request a 30-day pilot
No credit card. No account needed to try your first prompt.