Governance

Governance is the layer that makes an autonomous AI company safe to act. It is not a separate product — it wraps every executive decision.

​

Approvals

High-impact actions route through configurable approval workflows before they take effect. Thresholds are based on factors such as spend and risk:

• Low-impact actions proceed under the agent’s own authority.
• Higher-impact actions require sign-off from the parent C-Suite agent, the CFO, or — above a threshold — the Board.

When human approval is required, the agent prepares an evidence pack (context, analysis, recommendation, risks, alternatives) and waits. Nothing proceeds until it is approved.

​

The Sentinel veto

The Chief Risk Officer (AI_CRO, “Sentinel”) can veto any decision on risk or compliance grounds. The veto is deterministic and recorded; only the CEO can override it, and only with Board notification.

​

The audit trail

Every decision and sensitive action is written to a hash-chained, tenant-scoped audit log. The chain is verified on a schedule, so tampering is detectable. Each entry is attributable to a named agent role — which is what the EU AI Act’s human-oversight requirements expect.

​

Escalation

Work that exceeds an agent’s authority, crosses domains, or fails escalates up a defined chain rather than failing silently or quietly exceeding its mandate.

​

EU AI Act readiness

Operark is built as a high-risk AI system under Regulation (EU) 2024/1689. The platform maintains a technical file covering risk classification, prohibited- practice clearance, and the named-accountable-role mapping. High-risk obligations bind on 31 December 2027; Operark’s roadmap targets conformity assessment ahead of that date.