Skip to content
Cordum
Solution

AI Agent Governance for Regulated Industries

Out-of-process controls and audit evidence for regulated deployments. Cordum supports evidence collection; your auditor determines applicability and compliance.

An out-of-process decision point can support separation-of-duties controls. It gives reviewers policy evidence that does not rely only on the governed workload's own logs.

Cordum's Safety Kernel runs as a separate gRPC service and supports mTLS. Deploy it with separate identities, least-privilege storage access, and independent log access to establish the intended trust boundary.

That separation can support a regulated control program, but it does not clear an audit by itself. See the architectural deep dive and validate the deployment with your auditor.

Built for these verticals

Financial services

Transaction-limit policy, multi-party approval, and audit-evidence export for teams implementing financial-services controls.

Healthcare

Separation of duties between agent runtime and policy decision point, with tamper-evident records that can support a healthcare control program.

Public sector and regulated SaaS

Multi-tenant isolation and decision evidence for public-sector or regulated SaaS teams implementing their own regulatory control mappings.

Control questions to validate with your auditor

These architecture properties commonly shape audit discussions. Their sufficiency depends on deployment and control scope.

Trust boundary separation

The policy decision point runs outside the agent process. With least-privilege deployment and separate credentials, this creates an independently controlled evidence path.

Independent log stream

Policy decisions, approvals, state transitions, and evidence pointers are written to the control-plane store. Access controls determine whether reviewers can rely on it independently of workload logs.

Service identity and policy provenance

Safety Kernel clients support TLS/mTLS, and policy bundles can be signed for verification. Certificate, key-management, and log-access controls determine the evidentiary value of the resulting records.

Compliance evidence pack

A built-in SOC 2 control mapping provides a pragmatic starting point for evidence export. It is not an audit opinion, and your auditor must validate applicability.

Frequently Asked Questions

Why does out-of-process governance matter to my auditor?
An out-of-process policy service can support separation-of-duties and independent-evidence controls because it does not rely only on the governed workload's own logs. Your auditor must determine how that architecture maps to the controls in scope.
How does Cordum compare to Microsoft Agent Governance Toolkit for regulated buyers?
The relevant distinction is deployment topology: an in-process toolkit shares more of the governed workload's trust boundary, while Cordum can run a separate policy service with TLS/mTLS and control-plane evidence. Evaluate either approach against your threat model, tenancy model, deployment controls, and auditor requirements; no framework choice establishes compliance by itself. See the full comparison at /compare/cordum-vs-microsoft-agt.
What evidence does Cordum produce for an EU AI Act audit?
Cordum can export policy decisions, approval identities and timestamps, state transitions, and policy snapshots that may support EU AI Act evidence collection. Cordum does not certify compliance; legal counsel and your auditor must validate the applicable articles and evidence sufficiency.
Can Cordum run on customer-managed infrastructure?
Yes. Cordum is customer-managed and self-hosted today. The Safety Kernel, NATS, Redis, and workflow services run in infrastructure you operate. Contact us about deployment planning and rollout support.
How does CordClaw apply for OpenClaw deployments in regulated environments?
CordClaw is the OpenClaw governance adapter. On configured interception paths, it asks Cordum for a policy decision before execution and can record corresponding evidence. Coverage and bypass resistance depend on adapter configuration and endpoint controls, so validate them before a regulated rollout. See /edge/cordclaw for the adapter architecture.
What happens during a Safety Kernel outage?
Failure behavior depends on the configured mode and action path. Regulated deployments should validate fail-closed settings, timeouts, managed enforcement, alerts, and the documented outage runbook before rollout.

Compliance and audit reading

Practical guides to AI agent compliance frameworks, audit trail design, and policy enforcement evidence.

Talk to us about your audit

Bring us your auditor's questions — separation of duties, audit-trail tamper resistance, and evidence-export scope. We can walk through the documented trust boundaries, demonstrate configured failure behavior, and show how supported exporters send evidence to a SIEM. Your team and auditor determine control sufficiency.

Talk to security engineeringCompare vs Microsoft AGT