Appearance
From Zero to Audit-Ready
A complete end-to-end tutorial for achieving audit readiness with Modulos: from first scope to exportable audit artifacts.
In Modulos, audit readiness is not a last-minute document chase. It’s a continuous workflow: you establish a baseline, link the proof, and keep it current as the system evolves.
Not legal advice
This guide explains platform workflows. Your compliance obligations depend on your organization, jurisdiction, and framework interpretation.
The operating model
Audit readiness runs as a loop:
- Scope and map Set the boundaries of the AI System, apply the frameworks, and select the owners. Each Requirement Owner sets the Requirements that do not apply to Out of Scope.
- Implement Do the Control work, write the report of each Control, and attach the Evidence.
- Review Set the Control statuses. When all mapped Controls of a Requirement have a final status, Modulos notifies the Requirement Owner. The owner examines the Controls and marks the Requirement Fulfilled.
- Export and start again Export the AI System PDF. After a material change, go back to the scope and quantify the risks again.
This loop is driven by reality: a new model release, a vendor change, an incident, a data shift, or a framework update should trigger a refresh of scope, Controls, Evidence, and approvals.
What “audit-ready” means in practice
An audit-ready AI System typically has:
- a clearly defined scope (what system, what boundaries, what is not included)
- one or more frameworks applied, with applicability decisions recorded
- Requirements mapped to Controls (so you can show traceability)
- Controls in a final state and Requirements reviewed by their owners (audit trail)
- Evidence attached that is specific, attributable, and time-bounded
- key risks quantified, with exposure compared against risk limits
- optional but powerful: testing Sources and results linked to relevant Controls
1) Scope the system and the audit objective
Start by writing down the minimum scope statement you can defend:
- System: what is the AI system/product/use case?
- Boundaries: environments, regions, business units, integrations, models, and data flows in scope
- Stakeholders: owners, implementers, reviewers, and an “audit liaison”
- Objective: certification, regulatory readiness, procurement diligence, or internal governance
Best practice
Treat scope as a governance artifact. A well-scoped AI System reduces Control churn and makes “out of scope” defensible.
2) Choose the compliance lens: frameworks shape the work
Frameworks aren’t interchangeable. They define:
- what the unit of compliance is (system vs organization)
- what “good Evidence” looks like
- what kind of review/certification you are preparing for
Learn more:
- EU AI Act: product safety regime with conformity assessment and post-market monitoring
- ISO 42001 and ISO 27001: organization-level management systems with recurring audits and certification
- NIST AI RMF: a pragmatic risk management framework spanning org and system concerns
How these frameworks differ and why it matters
| Framework | Typical “unit” | What you’ll spend time proving | How to structure work in Modulos |
|---|---|---|---|
| EU AI Act | AI system / use case | Technical documentation, risk management, traceability, post‑market monitoring, and operational Controls | One AI System per deployed system or use case; link Controls to system-specific Evidence and monitoring |
| ISO 42001 / ISO 27001 | Organization | Management system: policies, roles, training, internal audits, continual improvement | The organization AI System for the management system, plus (optionally) a registered AI System for each high-risk system |
| NIST AI RMF | Both | Governance + measurement + management; demonstrates a functioning risk process | Use it to connect system Controls, risks, and ongoing measurements |
Other frameworks and customer Requirements vary widely in how they treat certification, audits, and Evidence. The goal is to be explicit about your interpretation and prove it with traceable artifacts.
3) Build the baseline: Requirements → Controls
Once frameworks are applied, work from Requirements down:
- Assign ownership: the AI System Owner sets the Requirement Owners (final accountability). Each Requirement Owner sets the Control Owners (implementation). See Requirements.
- Triage applicability: each Requirement Owner sets the Requirements that do not apply to Out of Scope. A comment gives the rationale.
- Check the mapped Controls: each Requirement comes with its Controls already mapped by the framework. Open Mapped Controls on a Requirement to see the work that it needs.
Controls vs Requirements: different statuses
Controls track execution (for example Not executed → Executed). Requirements track fulfillment (for example Not fulfilled → Fulfilled), marked by the Requirement Owner after reviewing the linked Controls once they reach a final state.
4) Connect context: Sources vs Connectors
At “zero”, most teams already have the Evidence, but it is scattered across tools. To speed up the first pass, connect that context and use Scout to find and summarize it. Scout reads documents, tickets, and repositories through your Connectors, and cloud, model, and trace data through the Sources that it can read.
Connectors and Sources are intentionally different:
- Sources are service accounts attached to an AI System. They power Runtime Inspection (metrics → tests → results) for that specific AI System.
- Connectors are user-level accounts connected to individual users (typically via OAuth). They are used to bring external context into Scout and follow user permissions.
Keep permissions tight
Treat integrations as compliance-relevant. Use least privilege, document access scope, and review connected accounts periodically.
5) Create audit-ready Evidence that is fast and defensible
The fastest path to “audit-ready” is to produce a first baseline that is:
- consistent (same claim structure across Controls)
- attributable (who approved, when, and why)
- time-bounded (what period the Evidence covers)
- reproducible (stable exports and versioned documents)
Use Scout to accelerate the first baseline
If Scout is enabled for your organization, use it to speed up drafting and gap discovery. Scout can reference your AI System context (Controls, Requirements, Evidence) and can use connected Connectors and Sources when available.
Example prompts:
- “For this Control, draft a first-pass report for our AI System and list the Evidence we should attach.”
- “Search our connected tools for candidate Evidence for this Control. Return the top 10 artifacts and why they matter.”
- “What’s missing for this Requirement to be fulfilled, and which Controls/Evidence should we create next?”
Capture the rationale, not just the artifact
Auditors often care about decision logic: why you chose an approach, who approved it, and how you verify it over time.
6) Review, approve, and lock the audit trail
Audit readiness depends on reviews. The sequence runs Controls first, then Requirements:
- Complete the Control work, and write the report. Record any “out of scope” or compensating Control rationale in the report.
- Set the status, for example Executed. A final status locks the report. Modulos records the status change in the Comments and Logs of the Control.
- Once a Requirement’s linked Controls are in a final state, the Requirement is ready for review: the Requirement Owner reviews the completed Controls and marks the Requirement as Fulfilled. This review at the Requirement level is the approval record.
This creates a defensible audit trail and reduces rework when scope or frameworks change.
7) Export an audit snapshot and stay continuously ready
When you need to share your posture internally or with auditors, generate an AI System snapshot/report export.
Then make audit readiness continuous:
| Cadence | What to review | Typical output |
|---|---|---|
| Weekly | Key system changes, new vendors, new models/data, incidents | New risks, updated Control narratives |
| Monthly | Evidence freshness, access reviews for Connectors/Sources | Refreshed Evidence on the affected Controls |
| Quarterly | Framework mapping and “out of scope” rationale | Re-validated scope and applicability decisions |
| Pre-audit | Final review pass + export | Audit snapshot aligned to the audit period |
Optional: quantify key risks in money
Risk quantification is only useful if it produces monetary impact, so you can prioritize mitigation and investment.
For example, quantification changes a qualitative risk into a value in money:
- Qualitative risk “Model bias incidents can occur in production.”
- quantify its threats
- Quantified risk An expected loss of €850,000 a year
Use a time basis of one year when you quantify a threat. Modulos reads a risk value as a loss for one year. For the methods, see Risk Quantification.
Why Modulos is designed as a system of action
Many governance programs fail because they treat compliance as a document repository problem. Modulos is designed to keep compliance and risk management operational:
- Workflows: owners, reviews, and status changes create a living audit trail
- Traceability: Requirements ↔ Controls ↔ Evidence and tests connect intent to proof and outcomes
- Agents: Scout answers questions and writes text that you can copy into a report, and it finds candidate Evidence across connected systems. The Control Assessment Agent finds gaps and stores its assessment when the run completes. Evidence Agent suggestions and Scout proposals take effect only when a person accepts them.
- Continuity: the same AI System becomes your ongoing operating model, not a one-off deliverable
Agents are most valuable in two places:
- From zero: turn scattered artifacts into structured, audit-ready narratives and well-associated Evidence faster.
- Over time: when the system changes, identify what needs to be updated and propose the next best actions.
Learn more: Scout, Evidence Agent, Control Assessment Agent, Risk Agent, Human in the Loop