Appearance
Data Handling & Access
This page tells you what data Modulos stores, who can see it, and how integrations get access.
For formal due diligence material, security questionnaires, and compliance reports, see the Modulos Trust Center.
What data Modulos stores
The data that Modulos stores includes:
- Governance records: AI Systems, frameworks, Requirements, Controls, statuses, assignments, and comments.
- Risk quantification data: monetary risk values, the inputs and outputs of each quantification, and the totals for each AI System and for the organization.
- Evidence: the files that users upload to the Evidence Library of an AI System, with a name and a description. Evidence attaches to Controls only.
- Comments and Logs: the comments and the log entries of Controls, Requirements, Assets, Policies, and AI Systems.
- Documents and records: Assets, Policies and their versions, and Vendors and their documents.
- Scout conversations: the chat history of each user.
- Integration secrets: the credentials of Sources and the tokens of Connectors. Modulos does not show a saved secret again.
Access control model
Modulos uses role-based access control at two levels:
- Organization: the settings and the users of the organization. Each user is an Organization Member. The other organization roles are Organization Admin, Risk Manager, Policy Manager, and Registration Manager.
- AI System: the governance work in one AI System. The AI System roles are Owner, Editor, and Viewer.
Organization Admins manage the settings and the users of the organization. Some organization roles also give access to all AI Systems:
- Organization Admins, Risk Managers, and Policy Managers have all the rights in each AI System of the organization, also without an AI System role.
- Organization Registration Managers can see each AI System and make the decision on its registration. They cannot change the work in it.
- Other users see only the AI Systems where they have an AI System role.
Related: User Management, AI System Settings
Auditability and accountability
Modulos keeps the history of the governance work:
- The Comments and Logs of an object record its changes, the user who made each change, and the time.
- Comments tell why a user made a change.
- The AI System PDF shows each Control with its status, its report, its Comments and Logs, and the names of its Evidence items. The PDF does not contain the Evidence files: users download them one by one.
Related: Audit Trail, Reports & Exports
Integrations and secrets
Each type of integration has its own identity:
- Sources: service accounts that belong to one AI System. Runtime Inspection tests and Scout use them for automation and monitoring.
- Connectors: the accounts of one user, which Scout uses only for that user. The Google Drive Connector holds Google's
drive.filescope: Scout can read only the files that the user selected in the Google Picker. The Picker does not let a user select a folder. - API tokens: user tokens for access to the Modulos API.
An Organization Admin turns on each Source type and each Connector type for the organization. Until then, nobody can use the type.
Protect secrets like passwords:
- Modulos shows the value of an API token one time. You can revoke a token at any time.
- Give Source credentials only the access that is necessary, and rotate them regularly.
- To remove a Connector, delete it in Modulos and revoke its access in the external system.
Related: Integrations
Privacy and content boundaries
Access to data follows the permission boundaries of each scope:
- Each Source belongs to one AI System and gives service-to-service access.
- A Connector has the permissions of the connected user in the external system.
- Only users with access to an AI System can see its Evidence and its governance records.
If you have strict data classification rules, set internal rules for the Evidence that users can upload and for how they redact sensitive content.
AI features and human review
The AI agents of Modulos change the governance record in these ways:
- Evidence Agent: it only suggests. A user creates or saves each Evidence item.
- Control Assessment Agent: it stores its readiness assessment when the run completes, with no review step. The assessment never changes the status of a Control.
- Risk Agent: a Quantified run becomes the current value of the threat, with no approval step. The run keeps its full reasoning, and a user can replace the value with a new run or a different method.
- Agent Tests: a failed run sets each mapped Executed Control to Not Executed, and each Fulfilled Requirement of these Controls to Not Fulfilled. A passed run never changes a status.
- Scout: its tools only read the Sources and Connectors, and Scout cannot change the connected systems. In Modulos, Scout creates a Draft AI System, a Metric Test, or an Agent Test only when a user accepts its proposal. It copies files from a Connector or Source to Evidence Creation only when a user accepts its Evidence proposal. It saves an answer as Evidence only when a user selects Save answer as Temporary Evidence.
Related: Human in the Loop, Code of Responsible AI
Customer responsibilities
Security and privacy are shared responsibilities. We recommend these practices:
- use dedicated service accounts for Sources
- keep API tokens in a secret manager
- give least-privilege access in external systems
- upload only Evidence that is correct for its audience, and redact it when necessary
Get help
If you think that there is a security problem, need a security package for a vendor review, or want guidance for your deployment, contact Support.