If your fraud pipeline uses embedding similarity, vectors planted in your store can steer the model. verify() reports whether a vector resolves to an object your pipeline registered. It is not a detector of small or adversarial modifications, or of malicious content entering through your own pipeline.
Only if your fraud pipeline uses vector embeddings. Tabular fraud stacks are out of scope.
02Why it matters
Fraud models depend on the data they receive.
Planted embeddings. Vectors written into the store outside your pipeline can shadow real fraud patterns, and similarity flags nothing. Synthetic behavior that enters through your own pipeline is registered like any other input: Mnemo attributes origin, not whether the behavior is genuine. Precedent poisoning. For fraud copilots, planted embeddings cause misleading precedent retrieval and silently shift analyst decisions. Cross-account contamination. On marketplace and platform fraud, one account's signals can leak into another's embedding space.
03What this does
A registration check under fraud decisions.
Every embedding your pipeline registers carries a vector_uid in your account. The fraud model calls verify() before similarity. Vectors that do not verify are excluded or flagged for review.
04Scope
What Mnemo does not do.
Does not replace your system
Does not score risk
Does not handle case management
Does not apply to tabular fraud stacks
Mnemo focuses on one layer: attaching identity to embeddings and verifying them when they are used.
Mnemo Verify checks whether a submitted vector resolves to an object registered to the caller's Mnemo account or namespace. verified=false is a safe outcome, not an error, and not proof that a vector is foreign. Verify does not establish the truth, authorship or ownership of the content behind a vector.
05Who this is for
Head of Fraud Engineering. VP Risk at fintech, marketplace, and e-commerce. Head of Trust and Safety. CRO at insurance carriers.
06Why teams adopt it
Customer-facing fraud decisioning under regulatory scrutiny.
Tenant-scoped verification for marketplace fraud signals.
Fraud copilots requiring precedent provenance.
Embedding-based behavioral fraud model validation under MRM.