Runtime evidence in the shape your DPO can hand over.

The three things every EU AI governance lead is quietly arguing about

1. Assured Workloads gives you the substrate. The agent's still flying without a flight recorder. Cloud Audit Logs show you the IAM, the data access, the policy denial — they don't show you the agent's reasoning, the tool calls it made, or the policy that blocked one.
2. VPC-SC enforces data residency. It doesn't enforce purpose limitation. GDPR Article 5(1)(b) is a runtime property of the workflow, not a property of the network perimeter.
3. The decision plane doesn't have a default story. Org Policy + VPC-SC + Assured Workloads handles the data plane. The agent's decision plane — the policy engine, the model-risk tier, the PII redaction, the kill switch, the residency check that fails closed — is the part that has to land before 2 August 2026, when the GPAI Code of Practice is enforceable and the AI Office expects evidence on demand.

What Regulus emits that your DPO can use

• Article 9 risk-management evidence. Every model invocation tagged with the resolved model-risk tier, the validation evidence pointer, and the framework citation (NIST AI RMF MANAGE-2.1).
• Article 10 data-governance evidence. PII redaction events with the matched pattern, the resolved jurisdiction, and the retention window.
• Article 50 transparency evidence. Auto-generated "this is an AI" disclosures attached to outbound agent messages in user-facing contexts.
• GPAI Code of Practice mapping. Profile eu-ai-act includes the GPAI sub-profile; commitments map to specific control IDs in the audit chain.
• EU AI Act Annex III scoping evidence. Which use-category your agent falls into, captured at scaffold time and tagged on every event.

What an EU supervisory authority will ask for

The questions you'll see in a real walkthrough, and where to point them in the Regulus audit envelope:

1. "Show me your Article 9 risk-management evidence." Filter the audit chain by framework_citations contains 'eu-ai-act:Article-9'. Export as signed evidence envelope.
2. "How do you enforce purpose limitation?" The policy plugin denies tool calls whose purpose claim doesn't match the agent's registered purpose. Show the DENY events in the chain.
3. "What happens when residency fails?" The residency plugin fails closed. Show the FAIL_CLOSED events; explain that no PII left the region.
4. "How do you evidence model risk?" Every model invocation carries model_tier. Tier-3 invocations require dual-control authorisation, surfaced through ADK's ToolConfirmation primitive.

Where to start

• EU AI Act profile page — 12-section breakdown of the regulation and what Regulus delivers.
• GDPR profile page — same shape, for Article 5(1)(b) purpose limitation and Article 22 automated decisioning.
• Coverage matrix — pick your active regulations, see which Regulus controls deliver which requirements.
• Article 9 in code — the cornerstone article →