DPO Radio

The hardest design decision around AI-generated evidence is the approval gate. Evidence that auto-approves on creation fails the inspection test — the regulator asks who approved it and "the AI did" is not an answer. Evidence that requires human approval but provides no preview surface is friction without value. The middle ground is structured: AI generates the draft with full grounding, the approver reviews the draft with the underlying citations visible, and the human approval transition is recorded.
The Forseti draft evidence approval surface is built around the middle ground. AI-generated evidence enters with a draft status. The approval queue surfaces every draft per record with the underlying citations, the workflow run that produced it, and the suggested approval action. The DPO or the audit lead approves per item; both transitions are recorded.
AI-generated evidence is created with a draft status. No tool, no MCP integration, no scheduled job can create approved evidence directly.
The approval queue surfaces draft evidence grouped by parent record (DPIA, dossier filing, vendor assessment, incident response). The reviewer walks the queue per record.
Each draft surfaces with the underlying citations, the workflow run that produced it, and the parent record context. The reviewer sees what the evidence asserts and how it grounds.
Approve transitions the evidence to approved with the operator and time recorded. Reject keeps the evidence in draft with the rejection rationale recorded.
The transition from draft to approved is enforced as a human action under every code path. No external integration, no scheduled job, no AI tool can transition evidence to approved.
Draft created, transition proposed, transition approved or rejected. The audit trail walks the lifecycle end to end.



The transition from draft to approved is enforced as a human action under every code path. No tool, no integration, no scheduled job can bypass.

Rejections persist in the audit trail with operator, time, and rationale. The reject record is not silently dropped.

The approval-queue surface ships with citations and workflow-run context per draft. The reviewer sees the grounding before the approval transition.

Use Verifying Quotes and Drafts to check a source or challenge a result. The review record captures sampled statements, changes and the reviewer decision. An accepted draft with no review record is shown as Accepted unreviewed.
Ready to see a draft regulator letter walked through the approval queue with citations and workflow-run context? Request a personalised demo.


Run a pilot and walk the approval queue daily for 30 days against your AI-generated draft evidence.

Talk to our team about how the approval queue fits your evidence governance.
No. The transition from draft to approved is enforced as a human action under every code path. No tool, no MCP integration, no scheduled job can bypass the gate.
The approval queue surfaces draft evidence grouped by parent record. The reviewer walks the queue per record.
The rejection persists in the audit trail with operator, time, and rationale. The evidence remains in draft state; the workflow can be re-run to produce a revised draft.
The standard approval is single-approver. Multi-approver review for sensitive evidence types runs through the platform's standard approval governance rather than the AI-evidence-specific queue.
Yes. All tiers — Light, Standard, Enterprise — have the draft evidence approval queue.