Coru
Book a demo

ADI Platform · Security

Guardrails in code, verified outside the model

Three mandatory filters, eight configurable, three severity levels and an auditable per-turn trail.

An instruction in a prompt is not a control

Asking the model not to say something is a preference, not a guarantee. A Risk team does not approve a control the controlled system itself can ignore.

So the filters run before and after the LLM, in deterministic code, and the model does not pick the final action: it proposes within a closed catalogue whose enums are validated outside.

The filters

FilterMode
Prompt injection Mandatory
PII exfiltration Mandatory
Crisis signals Mandatory
Topical focus 3 levels
Legal advice 3 levels
Medical advice 3 levels
Financial advice 3 levels
Offensive language 3 levels
Competitors and comparisons 3 levels
Commitments and promises 3 levels
Sensitive data in output 3 levels
Severity

Permissive, moderate and strict. The three mandatory filters cannot be disabled on any agent.

What Risk can reconstruct

Per-turn trail

What the agent said, on which prompt and knowledge base version, and which filter acted.

PII tokenised at the edge

Personal data never reaches the model provider; the audit log records the interaction without the data.

Versioned regression QA

Every behaviour change is tested against the approved case set.

Canary and rollback

Gradual percentage release, with a one-step return to the previous version.

Per-organisation isolation

Multi-tenancy with server-side enforcement, not client-side.

Deployment in your perimeter

Dedicated VPC or private network, whenever policy requires it.

Certifications

ISO 27001 PCI SOC 2

Bring your compliance checklist

It is the first conversation we have, not the last.

Request a demo

Try the platform your Risk team approves.

AI agents in production at regulated institutions across LatAm. Guardrails in code, deployment inside your network, and a first agent live in one to three weeks.

Guardrails in code

Deterministic filters before and after the LLM, three of them mandatory and non-disableable, with an auditable per-turn trail.

Full cycle, one approval

Origination, onboarding, transactions and collections on one platform. Compliance clearance happens once.

Weeks, not quarters

Sandbox on day one, voice on your SIP trunk in two to five days, integration included in the contract.

Shall we start?

Please use your work email.

By submitting you accept our privacy policy. We do not share your information with third parties.

Newsletter

Insights on AI in regulated operations, once a month.

Governance, architecture and what happens after go-live. No spam.

By subscribing you agree to receive communications from Coru.