Governance

Compliance that executes, not just documents.

AIOPS turns your AI policies into running code: durable governance workflows, policy gates that block non-compliant deployments, and human oversight with cryptographic accountability.

Governance workflow

A durable pipeline from intake to approval.

Every model runs the same governed path, orchestrated by a fault-tolerant workflow engine. If a step fails, it resumes — nothing is lost, nothing is skipped.

Intake & scan

Documentation is ingested with fail-closed PII scanning. Nothing unscanned enters the governance record.

Assess & classify

Deterministic control scoring, fairness metrics, SHAP explainability and EU AI Act risk-tier classification run automatically.

Policy gate

OPA evaluates the results against your published Rego policies. Failures block promotion — no exceptions without an override.

Sign & release

Approved models receive signed evidence and move to deployed status; monitors begin streaming immediately.

Human-in-the-loop overrides

When the business must ship despite a failed gate, an authorized reviewer can grant a signed override — scoped, justified and time-boxed. Overrides expire automatically and every one is recorded in the audit trail, so exceptions never become the rule.

Signed overrides · automatic expiry · full audit trail

Policy-as-code, reviewed like code

Compose policies in a visual builder, compile to OPA Rego, and publish through git pull requests. Your governance team reviews policy changes the same way your engineers review code — with diffs, approvals and history.

Visual builder · Rego compilation · git pull requests

EU AI Act classification

Every system is classified into the Act's risk tiers automatically, with obligations mapped per tier — prohibited, high-risk, limited and minimal — so your exposure is known before a regulator asks.

Automated risk tiers · obligation mapping

Auditor portal

Give external auditors read-only access to live evidence, control scores and the compliance ledger — no screenshots, no exported spreadsheets, no stale point-in-time binders.

Read-only access · live evidence · ledger view
Framework depth

Controls codified, not paraphrased.

Each framework requirement is implemented as an executable rule with clause-level traceability — the score you present to a regulator is reproducible from the same inputs, every time.

  • ISO/IEC 42001 — 88 controls with clause references
  • NIST AI RMF — 72 checks across all four functions
  • EU AI Act — risk classification and obligation mapping
  • SOC 2 — control evidence aligned to trust criteria
  • Extensible rules engine — add internal frameworks as code
# Deterministic control — same input, same score, forever
rule ISO42001_A_6_2_4 # "AI system impact assessment"
  requires:
    - impact_assessment.exists
    - impact_assessment.reviewed_within_days <= 365
    - affected_groups.documented
  score: Decimal # never float
  evidence: clause_ref, artifacts, timestamp

Put your policies in the deployment path.

See a policy gate block a non-compliant model live — and an authorized override resolve it with a signed, expiring exception.