EU AI Act & AI Agents
Compliance layer · EU AI Act

The registry of AI decisions that makes every verdict traceable

Every decision an agent produces via DAMM leaves an immutable trail: what was decided, with which scores, under which human oversight and with which transparency disclosure. An audit trail designed for the EU AI Act.

Three EU AI Act requirements, one registry

The DAMM registry is designed around the articles that concern the traceability of automated decisions.

Art. 12

Record-keeping

Every verdict produced via API/agents creates an immutable registry entry: input summary, scores, verdict, model and timestamp. An append-only trail designed for log record-keeping.

Art. 14

Human oversight

Each entry can be reviewed by a person who records the outcome — acknowledged, overridden or rejected — with a decision and notes. Human oversight becomes documented, not implicit.

Art. 50

Transparency

The registry records when the AI-generated disclosure (Art. 50) was served to the caller, with the exact text and time. Transparency toward the user becomes verifiable after the fact.

How it works

1 · The decision goes through the API

An agent or app asks DAMM to analyze a decision via the API. The engine produces scores across the 4 pillars, a Health Score and a verdict.

2 · Immutable snapshot

At the verdict, DAMM records a write-once snapshot — denormalized, so the trail survives even if the decision is deleted. No one can rewrite the scores after the fact.

3 · Oversight and export

From the dashboard a human records oversight (Art. 14), while the served disclosure (Art. 50) is tracked automatically. The whole registry is exportable to CSV for an audit.

What it records

  • Decision ID and source (API/agent)
  • 0-10 scores of the 4 pillars + Health Score
  • Verdict summary and risk level
  • Model, tokens and cost
  • Art. 50 disclosure served (text + date)
  • Oversight status, human decision and notes
immutable snapshot
{
  "decisionId": "3f7c1a2b-…",
  "source": "api_complete",
  "healthScore": 70,
  "model": "deepseek-chat",
  "verdictSummary": "…",
  "disclosureServed": true,      // Art. 50
  "oversightStatus": "pending",  // Art. 14
  "createdAt": "2026-08-01T10:05:00Z"
}

A registry entry, concretely

Here is how a single recorded decision looks: immutable verdict, Art. 50 disclosure and Art. 14 oversight in the same row.

Approve €4,200 out-of-policy customer refund

Decision ID: 3f7c1a2b-8e4d-… · api_complete · deepseek-chat

Medium riskHealth 58/100
Delimitation
7/10
Asymmetry
5/10
Margin
6/10
Minimum Move
5/10
Verdict

Proceed with caution: authorize a partial refund and check the customer history before the full settlement.

Art. 50 · Transparency served
01/08/2026 10:05 · AI-generated disclosure logged
Art. 14 · Oversight: Acknowledged
Human decision: approved partial refund €2,000
manager@azienda.com · 01/08/2026 11:20

Illustrative sample data.

For technical teams

  • Automatic population from the API layer
  • Read-only endpoint GET /registry
  • No latency impact (fire-and-forget)
  • Isolated sandbox: test keys leave no trace

For compliance & management

  • Append-only trail for log retention
  • Human-oversight record per decision
  • Auditor-ready CSV export
  • Explicit mapping to AI Act articles

Frequently asked questions

What is an AI decision registry?

It is a structured, immutable archive that stores, for each decision produced by an AI system, what was decided, with which scores, which model and when — together with the human oversight and the transparency disclosure. It makes automated decisions traceable and auditable.

Why is it useful for the EU AI Act?

It contributes to three EU AI Act requirements (Reg. EU 2024/1689): record-keeping (Art. 12), human oversight (Art. 14) and transparency toward users (Art. 50). It does not guarantee conformity, but provides the documentary trail these articles require.

Which decisions are recorded?

Only decisions produced via API/agents (complete and conversational modes). Test decisions (sandbox API keys) are not recorded, because they do not produce a real verdict and are not compliance-relevant.

Is the registry immutable?

Yes. The verdict snapshot is write-once: scores, Health Score and summary cannot be rewritten after creation. Only the human-oversight fields are updatable, as intended by the Art. 14 oversight principle.

Can I export the registry for an audit?

Yes. From the developer dashboard you can filter the entries and export them to CSV, ready to share with auditors or consultants. The registry is also queryable via a read-only API.

Does the registry replace legal advice?

No. The registry is a technical traceability and compliance-support tool. It does not constitute legal advice and does not replace a conformity assessment or the role of a DPO.

Make your agents’ decisions traceable

The registry populates itself from the API layer. You focus on oversight; DAMM keeps the trail.

This page is for informational purposes and does not constitute legal advice. The AI decision registry is a traceability tool that supports compliance with the EU AI Act (Reg. EU 2024/1689); it does not guarantee conformity and does not replace a conformity assessment or the role of a DPO.