Has AI policy on paper and wants to enforce it in practice now, not hope employees read it.
Compliance · Guardrails
Guardrails
Turn your AI policy into guardrails the control plane enforces deterministically. Numaga ships the defaults; you switch them on or off, or tune them for your organisation.
Who it’s for
Wants sensitive data and risky content caught at the gate, not discovered after the fact.
Wants to see when personal data would be shared, and have it masked or blocked automatically.
The problem
Policy no one enforces is an assumption.
You have written AI policy: which data may go to a model, what may not, which use is prohibited. But a policy document stops no one. Without enforcement, you only learn something went wrong once it already has, if you find out at all.
Guardrails turn that policy into behaviour on the control plane. Personal data, forbidden keywords, harmful content: all caught before the model call. And every violation that would otherwise have gone unseen now sits in your audit trail.
How it works
Numaga defaults, tuned for your organisation.
Guardrails are ready; you decide which are on and how strict.
Personal data: block or mask
The control plane recognises PII in every prompt. Per organisation you choose what happens (block, or mask automatically before the model) so GDPR-sensitive data never leaves the perimeter unchecked.
Your own keyword lists
Define which terms may not reach a model: project names, trade secrets, client names. The guardrail catches them at the gate.
Harmful content
Numaga defaults for self-harm and NSFW content are ready. Turn them on for your whole organisation or tune them per group.
Enforced and recorded
Every guardrail runs deterministically on the control plane, before each model call, and every hit sits in the immutable audit trail. Violations that would otherwise stay invisible now surface.
In practice
Proven where data is most sensitive.
Numaga is running a pilot with Ontdek Zorg, a home-care organisation in Gelderland. Healthcare, with special-category personal data under the GDPR, is where responsible AI is hardest. That is exactly where we put the control plane to the test.
- Home care in Gelderland
- Special-category data (GDPR Art. 9)
- Policy and redaction on every prompt
Put your AI policy into practice.
We show how to enforce your policy as guardrails on the control plane.
Request demo access