Built for risk, compliance, AI governance, and ML platform teams responsible for high-risk AI evidence.

Living evidence dossier · EU AI Act

Audit-ready AI evidence in minutes.

Guardian is the evidence and auditability layer above monitoring tools, review workflows, incident processes, system changes, and oversight activity—not another dashboard.

Capture activityLink evidenceExport audit-ready dossier

Book a pilot discussionExplore the workflow

Not ready to talk? Get the readiness checklist

RecordCustodyPolicy
Syncing sources

Guardian · Evidence record

EU AI Act evidence dossier

dcs_01842
BuildingSourcesPolicyIntegrityExport

Owner

—

Reviewer

—

Integrity

Computing…

Evidence trail

0/4 linked

SeqSource eventHandlerStatus
01model.eventguardian-ingestawaiting
02review.noterisk-committeeawaiting
03incident.openedoversight-queueawaiting
04oversight.logcompliance-deskawaiting
evidence-pack-dcs_01842.zip

Chapter II · Traceable evidence trail

How the evidence record is built.

A formal trace from production signal to register-backed export — each step logged, linked, and reviewable under scrutiny.

Custody logREF · GRD-COC-0847
SeqEvidence stepSource / ownerRecord detailStatusTimestamp (UTC)
001Signal capturedguardian-ingestthreshold breach · credit-risk-v3.2Captured2026-06-10 14:02:11
002Review assignedrisk-committeeoversight queue · manual reviewAssigned2026-06-10 14:01:58
003Policy trace linkedcompliance-deskEU AI Act · Annex III · Art. 9–11Linked2026-06-10 14:01:44
004Export readyrisk & complianceevidence-pack-dcs_01842.zipReady2026-06-10 14:02:18

All events linked · 4 entries · export authorised for inspection

Exhibit A

Operating view inside the evidence record.

Secondary reference only — the working surface behind the dossier, not the brand moment.

Appendix · Evidence workspace

Illustrative
Search decisions…
dcs_01842 · Credit scoring9:41 AM
Audit-ready
dcs_01839 · Hiring screen8:12 AM
Review needed
dcs_01836 · Claims AIYesterday
Evidence pending
dcs_01833 · Credit scoringYesterday
Audit-ready

Decision dcs_01842 · EU Credit Risk

eu-credit-riskmodel: credit-v3.2 · 9:41 AMArticle 9
Evidence summary

Decision inputs captured, model version linked, human review recorded. Evidence completeness: 92%. Policy trace linked. Export-ready.

  • Decision ID: dcs_01842
  • System: eu-credit-risk · Model: credit-v3.2
  • Risk score: 0.81 · Action: manual_review

evidence-bundle-dcs_01842.zip

Fig. 1Guardian evidence workspace — illustrative layout for governance review. Annotation: decision records, review state, and export paths within an existing dossier.

Inspection record · Appendix B

Hard questions under regulatory scrutiny

  • Decision trace

    Can you reconstruct an AI decision made 9 months ago?

  • Human oversight

    Can you prove human oversight actually occurred?

  • Regulatory scrutiny

    Would your current logs survive regulatory scrutiny?

Most AI teams cannot answer these questions months later. Guardian makes the evidence reconstructable, reviewable, and defensible.

Chapter III · Case files

High-risk systems under scrutiny.

Compact dossiers — what happened, what Guardian captured, what regulators can verify.

Illustrative examples — not customer case studies.

Case files registerREF · GRD-CF-2026
CaseWhat happenedEvidence capturedRegulator can verify
CF-01Hiring AIReadyAn ML model scored 847 candidates for a logistics role.Score rationale, feature weights, model version, review notes, and fairness drift signals captured at inference time.Annex III Article 9 conformity record, reviewer sign-off, and fairness audit trail.
CF-02Credit scoringLinkedA credit model declined 1,200 applications in Q3.Decision inputs, threshold state, override log, and model version reconstructed 9 months later.Full decision audit trail and human oversight record.
CF-03Healthcare triageLinkedAn AI triage system prioritized 340 patient cases.Clinical input snapshot, model version, confidence score, and escalation record locked at decision time.IVDR-aligned evidence bundle and accountability chain.

3 case files on record · register-backed exports available for inspection review

Appendix A · Source registry

Where the evidence comes from.

Guardian ingests signals from the systems you already run — no stack replacement.

View API documentationFull integration list
Source registryREF · GRD-SRC-REG
Source IDSystemChannelSignal typeEvidence role
SRC-01Python SDKAPImodel.metricsIngest · production metrics
SRC-02REST APIHTTPreview.eventCapture · review events
SRC-03WebhooksPushincident.openedAlert · incident intake
SRC-04Hugging FacePipelinemodel.versionLineage · model registry
SRC-05Azure MLPipelinedeployment.logTrace · deployment record
SRC-06AWS SageMakerPipelineendpoint.eventTrace · endpoint activity
SRC-07Weights & BiasesMonitordrift.metricSignal · drift observation
SRC-08DatadogMonitoralert.firedOps · runtime alert
SRC-09SnowflakeWarehousebatch.exportArchive · batch evidence

9 registered sources · normalized to evidence record · no stack replacement required

Chapter V · Audit preparation worksheet

Institutional readiness checklist.

A structured sprint to map evidence gaps, assign ownership, and produce one defensible baseline before portfolio expansion.

  1. Start with one high-risk system
  2. Map the evidence needed for oversight
  3. Assign ownership and review cadence
  4. Leave with a baseline your team can maintain

Chapter IV · Reading room

Editorial guidance for evidence-led teams.

Long-form notes on monitoring, incident registers, and post-market oversight — written for compliance and engineering leads.

RefTypeTitleExtent
RR-01GuideMonitor high-risk AI systemsHow to track metrics, events, and review states without rebuilding your MLOps stack.12 pp.RR-02FrameworkAI incident registerHow to structure incidents, owners, thresholds, and evidence for internal oversight.8 pp.RR-03ArticlePost-market monitoringWhat continuous monitoring means under the EU AI Act and how to keep records current.6 pp.

Common questions

Guardian in practice

What does Guardian do?
Guardian captures production signals, reviews, incidents, and oversight activity into one traceable evidence record—and packages it for audit and regulatory scrutiny.
Who is Guardian for?
Risk, compliance, AI governance, and ML platform teams accountable for high-risk AI systems under EU AI Act and internal oversight requirements.
Does Guardian replace existing MLOps or monitoring tools?
No. Guardian sits above the monitoring, ticketing, and review tools you already use. It links their outputs into a defensible dossier without replacing your stack.
What kind of evidence does Guardian produce?
Timestamped event logs, linked review notes, custody trails, policy mappings, and exportable audit-ready dossiers your team can retrieve on demand.
How does Guardian help with EU AI Act readiness?
It gives you a living record of monitoring, documentation, and oversight activity for high-risk systems—so you can show what happened, who reviewed it, and what was decided.

Pilot assurance register

Built for controlled pilot adoption

Guardian is designed for teams that need to evaluate AI governance evidence without replacing their existing stack or committing to a broad rollout on day one.

  1. Start with one AI system

    Pilot Guardian on a single high-risk AI system before wider adoption. Map evidence sources, preserve oversight activity, and export a first evidence dossier.

  2. Keep your existing stack

    Guardian sits above monitoring, MLOps, review, incident, document and ticketing systems. It is designed to connect evidence, not replace every operational tool.

  3. Export your evidence

    Evidence records, reviews, incidents, documents and supporting context should remain portable for internal review, audit preparation and vendor exit planning.

  4. Scope security early

    Deployment model, access control, data residency and security review requirements are discussed during pilot scoping.

  5. Clear compliance boundary

    Guardian supports evidence readiness and auditability workflows. It does not provide legal advice, certify compliance, or replace conformity assessment where required.

Map your evidence gapsDiscuss a controlled pilot

Closing statement

Start with one AI system. Leave with one defensible evidence record.

A focused pilot — no infrastructure overhaul, no slide-deck compliance theatre.

Book a pilot discussionRead the documentation →