Skip to content

Re-identification table & key management #2

Description

@AliakseiT

Re-identification table & key management (deferred, detailed)

The pseudonymization boundary produces store-safe text plus reid_material
(Presidio encrypt items) and reads the AES key from an env var. The mechanism
is reversible; the management of the re-id secret and mapping is undefined.

Requirements

  • Re-id table / key stays hospital-side, never in the engine store. Define the
    storage location and access control; enforce the invariant that reid_material
    and the key never touch runs/ or any engine-persisted artifact.
  • Per-project keys (one key per intended-use project, not a global key).
  • Key rotation + re-encryption strategy for retained pseudonymized corpora.
  • Escrow / recovery policy for the re-id key (who can re-identify, under what
    authorization; audit log of re-identification events).
  • Serialization of reid_material for durable hospital-side storage (Presidio
    OperatorResult items are in-memory today).
  • Consider consistent pseudonyms across documents/elements for the same entity
    (longitudinal linkage) — currently each call is independent.

Verify

  • Test that reid_material never appears in any store/ path (property test).
  • Round-trip re-identification with a rotated key fails closed with a clear error.

Why deferred

This is a safety/governance design decision (key custody, authorization, audit),
not a mechanical build — it needs a deliberate policy, like the recall decision.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions