Donna · AI systems portfolio

Lab 01

Privacy Boundary Lab

A local demonstration of deterministic identity protection before data crosses an AI service boundary.

LOCAL DETERMINISTIC DEMO · NO EXTERNAL AI

This is a portfolio-scale educational reconstruction using simplified recognizers and fictional data. It is not production-grade de-identification, anonymization or legal/compliance assurance.

Where identity lives

  1. Customer zone · raw identityTrusted input
  2. Customer zone · raw identityLocal detection
  3. Customer zone · raw identityStable tokens
  4. Outside · tokens onlyExternal AI boundaryDeterministic simulation
  5. Outside · tokens onlyToken-preserving response
  6. Customer zone · raw identityAuthorized restoration
Text equivalent: the trusted input, local detection and stable tokenization all happen inside the customer zone, where raw identities exist. Only tokens cross the external AI boundary, which in this lab is a deterministic simulation, not a live service. The response comes back with the tokens intact, and only an authorized step back inside the customer zone restores the original values.

Run the demonstration

  1. Trusted input
  2. Local entity detection
  3. Stable tokenization
  4. Outbound payload inspection
  5. Simulated external processing
  6. Authorized restoration

Status Load a fictional sample or type fictional text to begin.

Named outcome

NOT YET EVALUATED

Protect identifiers to evaluate the outbound payload.

A · Trusted customer zone

Original input and local detection

Stays in this page. Nothing is stored or sent.

Local detection results

Not analyzed yet.

    B · Outbound payload

    Exactly what would leave the customer zone

    
                

    Tokens are highlighted. Anything marked as residual is a possible raw identifier.

    C · Protected identity map

    Token-to-value mapping (stays in the customer zone)

    Collapsed by default. A production system would protect this mapping with stronger access controls, encryption, key management and auditing.

    No protected session.

    SIMULATED EXTERNAL RESULT

    Deterministic stand-in for an external AI service

    No network request and no model. A fixed rule set rewrites the tokenized payload and keeps every token exactly.

    
                

    Authorized restoration

    Restore inside the trusted zone

    This checkbox demonstrates an authorization boundary. It is not real authentication.

    
              

    Clearing removes this page's references to the input, payload and mapping and empties the fields. It is not cryptographic memory erasure.

    Test a failure

    Each scenario runs the same engine with fixed fictional input and shows which named outcome it produces, and why.

    Audit timeline

    Records what happened, never the protected values.

      No events yet.

      What production would additionally require

      This lab shows the shape of the boundary. A production deployment would additionally require:

      Browser-side pattern matching alone does not provide production privacy protection.