Day 6 of 1460 minExploit ArchitectHands-on lab

Day 6: Cross-Tenant Leakage: Filters, Namespaces and Caches

Taught by Mr Elite Updated

By the end of today you can

  • Compare single-index tenant metadata with per-tenant collections and name the failure mode of each
  • Test for filter omission and filter injection using two controlled tenants and unique markers
  • Explain and demonstrate how a shared embedding or response cache leaks content between users
  • Write a finding with a request sequence, minimal leaked content and a severity argument

Yesterday you proved that one account could or could not read another account's documents through a retrieval endpoint. Today the question moves one layer down. Even when the main query path checks the caller, something shared underneath may not: a filter that was never applied, a namespace that was never selected, or a cache that was keyed without the tenant.

These bugs are quiet. The vector search is correct, the authorization check passes, and yet User B receives text that only User A uploaded. I will show you where to look, how to prove it with harmless markers, and how to write it up so a developer cannot wave it away.

In this lesson

  • Three ways to separate tenants
  • Filter omission and filter injection
  • Shared caches: the quiet leak
  • Watch me do it: prove a leak with markers
  • Evidence that stands up in a report
  • Lab: cross-tenant cache leak

Premium course

$27 one-time, lifetime access

This day is part of the full course. Unlock it to read every lesson, do every mission and track your progress to the end.

Privileged members get every paid course included. See the plans