DGL®
Integrate real-time AI governance
DGL governs AI requests, responses, decisions and actions within connected applications and workflows in real time.
The integration supports request and response assessment, governance decisions, intervention, accountability and evidence as the workflow operates. The connected application or workflow enforces the governance outcome.
Initiation can come from any connected source
A governed event may originate from a human, an AI agent, an application, an API, an automated workflow or another system.
The source of an event and the person accountable for a decision are separate. An integration must supply the identity, authority, policy and relevant context required by its agreed configuration.
What the integration supplies
The integration supplies the request or proposed action, the AI response where applicable, relevant source context and the required identity and authority information.
The organisation defines and maintains its own policies, roles and authority limits. Those responsibilities remain with the organisation.
The agreed API contract determines the required fields and authentication. This overview describes the control relationship; it is not a substitute for the integration contract.
The governance sequence
Request Governance → Response Governance → Final Governance Decision → Evidence
Request Governance assesses the proposed event and its authority, policy and risk context. Response Governance assesses the AI response where applicable. The Final Governance Decision determines the operational outcome. Evidence records the assessment, decision and associated context.
HAO is one component within Response Governance. Response assessment findings inform the final decision; they do not constitute the whole DGL architecture.
What the connected workflow does
| Outcome | Required treatment |
|---|---|
| Proceed | Permit only the action and conditions authorised by the recorded decision. |
| Review | Hold the action for an authorised review and recorded decision. |
| Escalate | Hold the action for the required higher or specialist authority. |
| Withhold | Prevent the action from proceeding on this governance decision. Preserve the stated reason. |
The integration must handle unavailable or invalid assessments according to its agreed controls. A missing valid permission must not be interpreted as permission to execute.
Human intervention
Where required, an authorised reviewer examines the proposed action, its context and the assessment. Their decision, rationale and applicable policy reference are recorded.
An invalid assessment or a prohibited action is not converted into permission by an ordinary approval. The current review and override rules remain authoritative.
Evidence
Governance evidence connects the assessed event with its recorded outcome. Where a subsequent human decision is recorded, the record distinguishes that decision from the system’s assessment.
The agreed integration defines which records are available and how authorised people can access them.
Discuss integration
Discuss the proposed actions, authority boundaries, intervention requirements and evidence needs for your workflow.
julian@dglsystems.ai