Skip to content

Core Concepts ​

This page defines the building blocks you’ll see throughout Modulos and how they connect.

Concept map ​

The organization holds what all its AI Systems share, and each AI System holds its own work:

  • Organization Shared by all AI Systems
    • Users and roles The users and their organization roles
    • Settings The currency and the default language
    • Shared libraries The framework templates, the risk taxonomy, and the risk limits
  • Defaults and shared libraries
  • AI System The unit of work
    • User access The Owners, Editors, and Viewers of the AI System
    • Governance Frameworks, Requirements, Controls, and Evidence
    • Risk Risks and their threats, quantified in money
    • Runtime Inspection Sources, metrics, tests, and results

Mental model

Frameworks define what’s required. Controls define how you satisfy it. Evidence proves it. The Requirement Owner's review makes it auditable.

Organization ​

An organization is your workspace boundary in Modulos. It defines who can access the platform and sets global defaults.

Learn more: Organizations

Users, roles, and permissions ​

Modulos is designed for cross-functional work. Access is typically defined at two levels:

  • Organization: configuration and administration
  • AI System: day-to-day work on Controls, Evidence, risks, and testing

Best practice

Select the owners carefully. Only the Control Owner and the owners of the mapped Requirements can set the status of a Control, and only the Requirement Owner can mark a Requirement Fulfilled. This keeps separation of duties in the audit trail.

Learn more: User Management

AI Systems ​

An AI System is the unit of work in Modulos. It is registered in the AI Registry and typically corresponds to one deployed AI system or use case.

The AI Registry lists two types:

TypeWhen it appliesTypical output
AI Application (every AI System you register)Governing a specific AI system in its deployment contextControl implementation + Evidence, risk register, testing results
Organization (the organization AI System, for organization-wide Requirements)Governing organization-wide Requirements (policies, vendor posture, global Controls)Organization-level Evidence and reports

Organization-scope frameworks (OFF-x) attach to your organization AI System, listed with Type Organization in the AI Registry. The AI Registry does not create new organization AI Systems; if your organization does not have one yet, contact Modulos support before rolling out organization-scope frameworks.

Learn more: AI Systems

Frameworks ​

A framework represents a regulation, standard, or internal policy structure (for example EU AI Act, ISO 42001, NIST AI RMF). Frameworks are how you scope “what applies” to an AI System.

In Modulos, frameworks can be mapped so that one Control can contribute to multiple frameworks. This reduces duplication and keeps implementation consistent across compliance programs.

Learn more: Frameworks in Modulos

Requirements ​

Requirements are the "what" of a framework: the high-level obligations that come directly from the source document.

Framework-specific

Requirements represent articles, clauses, or chapters from the original regulation or standard. They preserve the structure and language of the source document for traceability.

Example: The EU AI Act Framework has a Requirement representing Article 10 of the Act, "Data and Data Governance".

Requirements typically:

  • map directly to articles, clauses, or chapters in the original framework document
  • group related obligations into a single trackable unit
  • map to one or more Controls that operationalize implementation
  • provide traceability for audit evidence

Learn more: Requirements

Controls ​

Controls are the "how": the concrete measures you implement to satisfy Requirements.

Designed for reuse

Unlike Requirements, Controls are not tied to a single framework. A well-designed Control can satisfy Requirements from multiple frameworks simultaneously, reducing duplicated work across compliance programs.

Example: Control "MCF-24 Data Design Choices" maps to relevant Requirements in multiple Frameworks, including the Article 10 Requirement of the EU AI Act.

A Control commonly includes:

  • Guidance (what “good” looks like and how to implement it)
  • Ownership (who is responsible)
  • A report/narrative (your explanation of implementation)
  • Linked Evidence (the supporting artifacts)
  • Status (changed directly, and logged in Comments and Logs)
  • Links to Policies and tests (Policies, Metric Tests, and Agent Tests)

Learn more: Controls

Evidence ​

Evidence is the proof. It can be a policy, screenshot, export, log sample, contract, test result, or any artifact that supports a Control’s implementation.

Evidence should answer:

  • What is this artifact?
  • Which Control(s) does it support?
  • What claim does it substantiate (and for which timeframe)?

Learn more: Evidence

Status, readiness, and review ​

In Modulos, status is concept-specific, and the path to a final status differs between Controls and Requirements.

Controls are completed directly. A Control tracks execution. Its owners, the Control Owner and the owner of each mapped Requirement, set the status on the page of the Control, or for several Controls in the Mapped Controls of a Requirement, with no reviewer approval. A Control has one of these statuses:

StatusMeaningWho sets it
Not ExecutedThe work is open. A new Control starts here, unless the EU AI Act scoping sets it to Out of Scope.The owners. Modulos sets an Executed Control back to Not Executed when a linked test fails, a linked Metric Test has an error, or a linked Policy is archived.
ExecutedThe work is done, and its Evidence is locked.The owners
Out of ScopeThe Control does not apply.The owners, or the EU AI Act scoping
ArchivedA framework update removed the Control.Modulos

Executed, Out of Scope, and Archived are final statuses. Modulos records each status change in the Comments and Logs of the Control, with the user and the time.

A comment that tells why is optional. Modulos asks for a reason in one case only: an Executed Control goes back to Not Executed while a mapped Requirement is Not Fulfilled and ready for review.

Reviews happen at the Requirement level. A Requirement tracks fulfillment. It has one of these statuses:

StatusMeaningWho sets it
Not FulfilledThe Requirement is open. A new Requirement starts here, unless the EU AI Act scoping sets it to Out of Scope.The Requirement Owner. Modulos sets a Fulfilled Requirement back to Not Fulfilled when it sets a mapped Control back to Not Executed.
FulfilledThe Requirement Owner examined the mapped Controls and accepts them.The Requirement Owner, after all mapped Controls have a final status
Out of ScopeThe Requirement does not apply.The Requirement Owner, or the EU AI Act scoping
ArchivedA framework update removed the Requirement.Modulos

Readiness is a different signal. When all mapped Controls of a Requirement have a final status, the Requirement is ready for review. Modulos notifies the Requirement Owner, and a Not Fulfilled Requirement shows Mark as Fulfilled. While a Requirement is Fulfilled, no user can change the status of its mapped Controls.

Use “Out of scope” sparingly

Treat “out of scope” as a decision that may be challenged. Record the rationale in the Control/report so the audit trail is complete.

Learn more: Reviews & Statuses

Risk ​

Modulos supports risk management at the organization level and per AI System.

  • Organization taxonomy: the shared catalog of categories, risks, and threats your organization uses for consistency.
  • AI System risks: the risks that apply to an AI System, broken down into threats for quantification.

Learn more: Risk

Risk quantification ​

Risk quantification turns qualitative risk statements into monetary impact so you can prioritize mitigation and investment.

Why monetary?

5×5 risk matrices are not quantification and are often misleading.

Money is the point because it:

  • is a universal unit across risk types
  • forces abstract harms (reputation, trust, mission delay) into stakeholder-relevant terms
  • connects directly to organizational risk appetite, which is ultimately expressed through resource allocation

You can compare a monetary risk value with the value that an AI System gives. The Value and Risk view puts the annual economic value of each AI System next to its risk value, and shows which AI Systems give more value than risk.

Learn more: Risk Quantification and Value and Risk

Runtime Inspection ​

Runtime Inspection lets you continuously verify key properties of your AI system (or its operational environment) using metrics and rules.

The runtime inspection model is:

  • Sources connect to systems that provide metrics and context (for example Prometheus and Datadog for pull metrics, Modulos Client for pushed metrics, and GitHub and Azure for AI System context).
  • Metrics are the measurements available from a Source.
  • Tests define conditions over metrics (thresholds, comparisons, schedules).
  • Results are the historical outcomes you can reference in Controls and audits.

Learn more: Runtime Inspection

Connectors vs Sources

Sources are service accounts attached to an AI System, used to power Runtime Inspection (metrics → tests → results) for that AI System. Connectors are user-level accounts connected to a user (typically via OAuth) and used to bring external context into Scout.

AI agents and automation ​

Modulos includes AI agents to accelerate documentation and investigation workflows.

  • Scout: conversational assistant grounded in your AI System context (Controls, Evidence, Requirements).
  • Evidence Agent: suggests the Controls for uploaded Evidence, and in the Evidence dialogs a name and a description. Nothing becomes Evidence until a user creates or saves it.
  • Control Assessment Agent: assesses the readiness of a Control from its Evidence and linked Policies. It stores the result with gaps and recommendations, and it never changes the status of the Control.
  • Risk Agent: autonomously quantifies a risk threat in monetary terms, recording the full reasoning trace; runs can be repeated or replaced with another quantification method.

Learn more: AI

Reporting and audit readiness ​

Audit readiness is an outcome of consistent traceability:

  • Requirements → Controls → Evidence → reviews → exports

Learn more: Reports & Exports

Next steps ​