EU AI Assurance OSOpen Dashboard

Ship AI systems in the EU with evidence, not guesswork.

EU AI Assurance OS turns risk classification, evidence, eval gates, and data-contract checks into a single release decision — PASS, REVIEW, or BLOCKED — backed by an audit-ready evidence pack.

Open DashboardSee how it works
Claims Triage AI — Release Gate
BLOCKED
Evidence coverage: 6 documents indexed
Eval score: 0.93 (threshold 0.85)
Data contract: open BREACH on claims-intake-v2
Human oversight SOP: on file

Releasing AI in the EU shouldn't mean a fire drill

Evidence chasing before every audit

DPIAs, model cards, and vendor docs live in shared drives and inboxes, scattered across teams — until someone has to find them all at once.

Unclear obligations per risk tier

Each risk classification implies a different set of controls, but tracking which apply to which system is manual and easy to get wrong.

No single release decision

Eval scores, contract drift, and approvals live in different tools, so “is this safe to ship?” doesn't have one clear answer.

Everything a release decision needs, in one place

AI System Registry
Record every AI system with owner, purpose, risk class, deployment context, vendor/model info, data sources, and release status in one place.
Risk Classification
A guided workflow records risk class, rationale, affected users, sector, decision impact, and the controls each tier requires.
Evidence RAG
Get cited answers from policy docs, DPIAs, model cards, vendor docs, incident records, and data contracts — no more digging through shared drives.
Eval Gates
Store eval datasets, model and prompt versions, scores, thresholds, and guard metrics that feed directly into the release decision.
Data Contract Monitor
Track input schemas, semantic contracts, lineage, and drift events — with severity and remediation state visible before release.
Approval Workflow & Audit Ledger
Route blocked or in-review systems through owner, compliance, legal, and human-oversight approvals, with an append-only audit trail of every decision.

From registration to release decision

  1. 1

    Register the system

    Add the AI system with its owner, purpose, deployment context, and data sources.

  2. 2

    Classify risk & attach evidence

    Record the risk tier and rationale, then attach DPIAs, model cards, vendor docs, and policy evidence.

  3. 3

    Run eval gates & check contracts

    Score the model against faithfulness, bias, accuracy, and cost thresholds, and confirm data contracts have no open drift.

  4. 4

    Get a release decision

    Receive a PASS, REVIEW, or BLOCKED decision with the controls behind it, and export an evidence pack for audit.

Built for everyone in the release path

AI Engineering Lead
Get release gates for model and prompt changes, wired to eval scores and thresholds you control.
Compliance Officer
See evidence that every AI system meets its regulatory obligations, with citations you can hand to an auditor.
Legal Counsel
Get documented risk classifications, oversight records, and audit trails for every release decision.
Data Platform Lead
Monitor data-contract drift and lineage so a broken upstream schema can't silently block or sink a release.
Product Owner
Get a clear PASS, REVIEW, or BLOCKED decision before launch — with the reasons, not just a status.
Auditor
Review an immutable record of who approved what, when, and on what evidence.
EU AI Act-aligned controlsTenant data isolationEncryption in transit & at restAppend-only audit ledgerDeterministic evidence export

Frequently asked questions

What is an EU AI Act risk classification?

It's the tier (e.g. minimal, limited, high) assigned to an AI system based on its sector, decision impact, and affected users. The tier determines which controls and evidence are required before release.

What is an evidence pack?

An evidence pack is a deterministic, exportable bundle of the documents, citations, eval results, and approvals behind a release decision — built for audit review.

How does an eval gate determine release readiness?

Each AI system has eval runs scored against thresholds for faithfulness, bias, refusal behavior, accuracy, latency, and cost. The latest completed run must meet its threshold for the release gate to pass.

What counts as a data contract drift event?

A drift event is recorded when an input data source no longer matches its agreed schema or semantic contract. An open breach-severity drift event blocks the release gate until it's resolved.

Who needs to approve a high-risk AI system release?

High-risk systems route through owner, compliance, and legal approval, and require documented human-oversight evidence before the release gate can pass.

Does this replace a legal determination of EU AI Act obligations?

No. EU AI Assurance OS is a control plane that organizes evidence, evals, and approvals against EU AI Act-style obligations — it doesn't provide legal certification or a final legal determination.

Ready to see your release gate?

Open the dashboard to register a system, run an eval gate, and get your first PASS, REVIEW, or BLOCKED decision.

Open Dashboard