Which prompt filtering layer supports tenant isolation and strong false positive control?

The prompt filtering layer that best supports tenant isolation and strong false positive control is typically the tenant-specific policy/rules layer in the prompt filtering stack.

If you’re choosing among common layers:

• Pre-filter / policy enforcement layer: Best place for tenant isolation, because rules can be scoped per tenant before the prompt reaches the model.
• Deterministic rules / allowlist-denylist layer: Best for strong false positive control, since explicit rules are usually more predictable than ML-based filters.
• ML-based moderation layer: Better for broad coverage, but usually weaker on false positive control and harder to isolate cleanly by tenant.

So the short answer is: a tenant-scoped deterministic policy layer.