What is tenancy isolation?
Tenancy isolation is the separation of one organisation's data and compute from every other organisation's. Enforcement belongs to the boundary the infrastructure provides rather than to application logic above it. Isolation at that level is what allows an organisation to state where its data resides and who could reach it.
Also called: Tenant isolation
Why a predicate in the query path is a weaker boundary than it looks
The breach pattern in multi tenant systems is almost never a broken cipher. It is a query that omits the tenant predicate, or a cache key that collides between two clients. Each of those defects lives in application code, which is precisely why isolation enforced in application code fails. A bank asking whether its environment is shared is asking where the boundary lies, and the answer has to name a piece of infrastructure.
- A boundary below the application cannot be defeated by an application defect.
- Dedicated infrastructure makes the data residency answer verifiable per client.
- Isolation determines what a single compromised credential is able to reach.
- Regulated procurement treats shared compute as a finding requiring justification.
How tenancy isolation works
- The isolation model is chosen per client against the regime governing their data.
- Storage and compute are provisioned in a dedicated account or namespace.
- Network policy denies traffic between tenancies by default.
- Encryption keys and identities are scoped to a single tenancy.
- Vector indexes and caches are provisioned per tenancy rather than partitioned by a key.
- Isolation is verified by attempting cross tenancy access from inside the environment.
How tenancy isolation is evaluated
Isolation is demonstrated by attempting to breach it rather than by reading the architecture. Test the following from inside each tenancy:
- result of a cross tenancy access attempt holding valid credentials
- separation of encryption keys and of key administration
- scope of every shared cache and queue alongside any other shared component
- residency of each data store against the applicable requirement
- blast radius of one compromised service credential
Limitations, risks and common failure modes
- A vector index is shared across clients because the embedding cost appeared to justify it.
- Log aggregation collects content from every tenancy into one searchable store.
- An administrative console reaches all tenancies through a single credential.
- Isolation holds in the data path and leaks through a monitoring tool.
- A restored backup lands in the wrong tenancy during an incident.