Secure AI Collaboration: Zero-Knowledge PII Session Handoff
Share Sanitized AI Prompts & Reversible Token Maps Across Teams Without Cloud Databases (Argon2id + XChaCha20-Poly1305)
AI Summary / Key Takeaways
"Collaborate on ChatGPT and Claude outputs without exposing PII. Share reversible in-memory token maps via zero-knowledge links or .pssession files."
Enterprise-Grade AI Privacy
Add custom redaction rules and priority support with PRO.
Interactive PII Detection & Sanitization Sandbox
Test real-time client-side RAM tokenization. Choose a specialized preset or paste your own raw prompt to test instant reversible redaction.
Sanitized locally in volatile RAM. Pre-prompt compliance maintained under Secure Team Handoff standards (GDPR, HIPAA, SOC 2).
Why Shared Cloud AI Workspaces & Standard Encryption Break Down
When an analyst masks a 20-page contract before prompting Claude, the AI response comes back filled with placeholders like [NAME_1] and [ESCROW_2]. Without the local RAM session map, the reviewing partner is stuck: either waste 25 minutes manually re-typing real names from a spreadsheet, or dump the raw mapping table into a shared cloud drive that creates a centralized breach liability. Standard end-to-end encryption (PGP) cannot solve this because an LLM cannot analyze encrypted ciphertext. Teams need a way to let cloud AI reason over sanitized structure while transferring the decryption state peer-to-peer.
How Dual-Plane Session Handoff Works
Plane 1: Local RAM Tokenization for Cloud AI
Sanitize contracts, clinical records, or financial models in browser RAM (<2ms). Send only the structural cleartext with surrogate tokens ([NAME_1], [FINANCIAL_1]) to ChatGPT or Claude so the LLM can reason accurately without seeing PII.
Plane 2: Protocol V5 Zero-Knowledge Packaging
Your browser compresses the volatile surrogate mapping table via in-memory stream compression, derives a memory-hard key with Argon2id, and encrypts it with XChaCha20-Poly1305 into a compact #hash Magic Link (<1,850 chars) or .pssession file.
0.4-Second Recipient Reveal (Free for Recipients)
Your colleague, reviewing partner, or external client opens the #hash link (which never touches any HTTP server) and restores every original name and figure inside the AI response in 1 click—with zero account or seat license required.
Supported Formats & Limitations
Supported Formats
System Limitations
- Requires out-of-band shared passphrase for Argon2id decryption
- Direct #hash links cap at 1,850 base64 characters (larger maps auto-export as .pssession files)
- Zero-knowledge architecture: lost passphrases cannot be reset or recovered by PrivacyScrubber
Under Zero-Trust Data Sanitization (ZTDS™), collaborative AI workflows must never rely on a centralized cloud token database or vendor key vault. PrivacyScrubber separates AI reasoning from identity restoration using Protocol V5 (Argon2id + XChaCha20-Poly1305), establishing a verifiable Statutory Risk Minimization & Technical Safe Harbor under GDPR Art. 32, HIPAA §164.312(e), and Israel PPL Amendment 13 Sec. 17.
Stop wasting 25 minutes per contract manually replacing [NAME_1] and [ESCROW_2] from a side-by-side spreadsheet. Your colleague restores 100+ entities across a 15-page Claude output in one click.
Only the sender needs a TEAMS license. Outside counsel, clients, auditors, and external partners open your #hash link and reveal restored AI outputs in local RAM for free—no account, no login, no paywall.
Enterprise cloud DLP gateways charge $35–$60 per seat every month. Because ZTDS™ runs 100% on client CPUs with zero server database overhead, one $99/mo license covers your entire department.
Why Standard Encryption (PGP, Signal, Password Links) Fails in AI Workflows
Security architects and legal counsel evaluating collaborative AI tools frequently ask a fundamental question: "If we encrypt a payload with a passphrase and send a link, how is that different from PGP, Bitwarden Send, or an encrypted ZIP file?"
The answer lies in The AI Encryption Paradox: standard end-to-end encryption (E2EE) is designed for static human-to-human message transport, which completely breaks Large Language Model reasoning. If you encrypt a contract with PGP before sending it to Claude or ChatGPT, the AI sees random ciphertext bytes and cannot analyze clauses, calculate risk, or summarize obligations.
Conversely, if you use a traditional Cloud DLP Gateway that masks PII on a central server and stores the mapping table in a vendor cloud database, you create a centralized honeypot that triggers third-party Data Processing Agreements (DPAs), sub-processor audits, and cross-border transfer liabilities.
The ZTDS™ Dual-Plane Split Architecture
PrivacyScrubber resolves the AI Encryption Paradox by splitting collaborative AI work into two orthogonal, cryptographically isolated planes:
Your browser strips direct PII and domain identifiers in local RAM, replacing them with deterministic, typed surrogates ([NAME_1], [COMPANY_2], [FINANCIAL_1]). The cloud LLM receives clean syntactic context and performs full legal, clinical, or financial reasoning without ever touching raw personal data.
Instead of encrypting the entire document, Protocol V5 compresses only the volatile bijective token map (local ZTDS™ token dictionary) in client RAM, derives a 256-bit key via Argon2id, and seals it with XChaCha20-Poly1305 into a client-only URL #fragment (<1,850 chars) or .pssession file for instant colleague Reveal.
The Trust Topology: Zero Cloud in the Loop
During a Protocol V5 Session Handoff, the URL fragment (#...) is governed by RFC 3986 client-side isolation: browsers never transmit URL fragments in HTTP requests to any server, proxy, or CDN. The moment your colleague opens the link, PrivacyScrubber hydrates the ciphertext into volatile RAM and executes history.replaceState to wipe the fragment from the address bar.
2. Seals token dictionary (0x05)
Zero cleartext PII
auto-wipes URL hash
Architecture Comparison: ZTDS™ Handoff vs. Legacy Alternatives
Evaluating how to share AI-assisted work across legal, clinical, or financial teams? Compare how four common approaches handle LLM reasoning utility, server exposure, and operational speed:
| Collaboration Method | Can Cloud AI Reason Over Document? | Server / Cloud Access to PII Map? | External Recipient Friction | Re-Identification Time |
|---|---|---|---|---|
| ZTDS™ Session Handoff (Protocol V5) | Yes (Tokenized Structure) | Never (100% Client RAM) | Free 1-Click #hash Reveal | 0.4 seconds |
| Standard E2EE (PGP / Bitwarden Send / ZIP) | No (AI cannot read ciphertext) | No (Static file only) | Requires manual token splicing | 20+ min manual |
| Cloud DLP Gateways & Central Token Vaults | Yes | Vendor server stores DB | Paid seat + SSO required ($40+/u) | Proxy latency |
| Shared Google Doc / Notion Token Key | Yes | Cloud provider indexes PII | Shared drive permissions | 15–25 min/doc |
Statutory Risk Minimization & Technical Safe Harbor
Free-text AI prompts can contain contextual narrative details, meaning no technical control should be misrepresented as a blanket legal exemption from data protection statutes. Instead, ZTDS™ Session Handoff is engineered to provide Statutory Risk Minimization and a Technical Safe Harbor (Defense-in-Depth)—demonstrating proactive, state-of-the-art organizational due diligence that shields DPOs, CISOs, and corporate directors from personal administrative liability:
| Regulatory Standard | Statutory Provision | Technical Safe Harbor Implementation |
|---|---|---|
| GDPR (EU / UK) | Art. 25, Art. 32 & Recital 26 | Enforces client-side pseudonymization and state-of-the-art cryptographic isolation; decryption keys never reside with the AI processor or cloud vendor. |
| Israel Privacy Law | Amendment 13, Section 17 | Establishes documented technical risk minimization and DPO governance controls, mitigating statutory administrative fines and officer liability. |
| HIPAA | §164.312(e)(1) & §164.514(b) | All 18 Safe Harbor PHI identifiers are tokenized in RAM prior to LLM transit; peer-to-peer handoff payloads are sealed with XChaCha20-Poly1305 AEAD. |
| SOC 2 Type II | CC6.1 & CC6.7 | Eliminates centralized credential stores and third-party database attack surfaces; verifiable via SHA-256 Compliance Receipts. |
| NIST & ISO 27001 | SP 800-132 & Annex A.8.24 | Memory-hard Argon2id key derivation paired with 256-bit XChaCha20-Poly1305 authenticated encryption and 192-bit random nonces. |
Reference Architecture Workflows by Sector
M&A & Outside Counsel
An associate scrubs counterparty names, cap tables, and escrow figures under the Legal Profile before prompting Claude. They send the AI redline plus a #hash Handoff Link to external co-counsel, who reveals the exact deal terms in 0.4 seconds without needing a paid seat.
Clinical & Utilization Review
Medical coders mask patient names, MRNs, and dates of birth before running clinical notes through AI for ICD-10 coding. The attending physician clicks the zero-knowledge link to verify the AI's diagnostic summary with patient identities restored locally.
FinTech & SecOps Triage
L1 security analysts scrub customer IBANs, JWTs, and production hostnames from incident logs before asking ChatGPT for root-cause analysis. L3 engineers import the .pssession package to inspect the AI remediation plan mapped to real production hosts.
Zero-Server Fleet Blueprints & 1-Click CISO Receipts
Beyond peer-to-peer session handoffs, the TEAMS tier equips CISOs and DPOs with zero-infrastructure departmental governance:
Export an AES-256-GCM encrypted Blueprint token or JSON policy package to enforce mandatory 30 Industry Detection Profiles and locked Custom Regex Rules across every browser seat with zero server configuration.
Generate client-side verifiable PSR-2 Compliance Receipts featuring content-derived SHA-256 digests, UTC timestamps, and framework mappings for SOC 2, HIPAA, GDPR Art. 32, and Amendment 13 audits.
$99/Month Flat · Unlimited Seats · Zero Per-User Tax
Whether your legal, clinical, or engineering team has 3 seats or 300, the price never scales with headcount. External clients and outside counsel decrypt handoff links for free. Need air-gapped on-premise source licensing? Explore our Enterprise Appliance tier.
Client-Side ZTDS™ Execution & Conformance Verification
All PII detection and deterministic tokenization execute 100% inside local client memory — eliminating cloud proxies, external subprocessors, and network data egress.
Frequently Asked Questions
Technical Details & Privacy Architecture
How does ZTDS™ Session Handoff differ from PGP encryption or password-protected links like Bitwarden Send?
Do external recipients, clients, or outside counsel need a paid TEAMS license to open a handoff?
Why doesn't the Direct Zero-Knowledge Link (#hash) leak the encrypted payload to web servers?
What cryptographic primitives power Protocol V5 Session Handoff?
How does Session Handoff support GDPR Article 32 and Israel Amendment 13 Section 17 compliance?
How many team members can use Team Handoff under the $99/month plan?
What happens if someone intercepts the #hash link or .pssession file in Slack or Email?
Share Sanitized AI Sessions Across Your Team in 0.4 Seconds
External recipients decrypt #hash links for free in local RAM. Unlimited departmental seats for $99/month flat.
