Managing Attribute-Based Access Control (ABAC) for AI Governance

Overview

In ABAC Phase 1: Foundations, Attribute-Based Access Control (ABAC) for AI Governance enables organizations to grant access to specific AI Governance records without providing broad access across the entire solution. By combining role-based access with record-level assignments, organizations can support collaboration across AI governance workflows while maintaining least-privilege access and auditability.

ABAC is particularly valuable in AI Governance because AI initiatives often require participation from multiple stakeholders, including governance teams, legal, privacy, security, architecture, and technical teams, who may each require different levels of access to the same AI record.

Important:

OneTrust is continuing to enhance the ABAC experience for AI Governance to deliver the intended customer experience. While customers are encouraged to begin configuring assignments and planning their implementation, ABAC functionality cannot yet be fully validated through end-to-end testing. Future release notes will provide updates when full testing support becomes available.

How ABAC Works in AI Governance

ABAC lets administrators manage access at the record level through assignments. This enables access to specific records, rather than broad access across AI Governance. In AI Governance, records include Projects, AI Systems, AI Agents, Models, and Datasets.

ABAC is primarily used to support controlled collaboration around specific AI initiatives. For example:

An AI Platform Owner can receive edit access to a specific AI System or AI Agent during intake, remediation, or governance reviews.
Legal, privacy, security, and architecture reviewers can receive view-only access to a sensitive AI record without requiring broader solution access.
Governance leads can share individual records with reviewers who need to participate in a decision or approval process.
Users can request access directly from a record when they need to participate in a review, investigation, or governance workflow.

The goal is to enable faster collaboration across distributed stakeholders while preserving least-privilege access and maintaining visibility into who has access to a record and why.

Common Use Cases

Use cases describe the most common business scenarios where ABAC helps teams collaborate on records while maintaining least-privilege access. The examples below highlight how organizations can use ABAC to provide the right users with access to the right records at the right time.


Use Case	How ABAC Helps
Cross-Functional AI Reviews	Enables legal, privacy, security, governance, and technical stakeholders to collaborate on a specific AI initiative without granting broader solution access. Example personas include: Head of AI Governance, AI Governance program owners and managers, legal reviewers, privacy reviewers, security reviewers, architecture reviewers, and model risk reviewers.
AI System and Model Governance	Allows owners and contributors to manage and update specific AI Systems, AI Models, AI Agents, or Datasets while restricting access to other records. Example personas include: AI Platform Owners, AI Governance program owners and managers, business owners, and technical contributors responsible for populating, updating, or reviewing AI System, AI Model, AI Agent, and Dataset records.
Point-of-Need Access Requests	Allows users to request access directly from records when they need to participate in reviews, approvals, or remediation activities. Example personas include: legal, privacy, security, and architecture reviewers, business owners, and technical contributors who need time-bound access to a specific AI record to participate in a review, approval, investigation, or remediation activity.
Review and Approval Processes	Provides stakeholders with the access necessary to review, evaluate, and approve AI initiatives while maintaining clear ownership and accountability. Example personas include: Head of AI Governance, AI Governance program owners and managers, governance leads, business owners, and legal, privacy, security, architecture, and model risk reviewers involved in evaluating and approving AI initiatives.
Least-Privilege Collaboration	Reduces custom role sprawl by granting access only to the records users need to perform their responsibilities. Example personas include: Chief AI Officers, Heads of AI Governance, AI Governance program owners and managers, AI Platform Owners, and cross-functional reviewers from legal, privacy, security, architecture, and model risk teams.

Key Workflows Impacted by ABAC

ABAC introduces record-level access controls within existing solution workflows. The following workflows are commonly impacted by ABAC and may change how users access, review, and collaborate on records.


Workflow	How ABAC Changes the Experience
AI Inventory Management	Enables record-level access to AI Systems, AI Agents, AI Models, Datasets, and Projects without modifying module-wide roles.
AI Intake Processes	Allows stakeholders involved in intake and review activities to access only the relevant AI records.
Assessments	Provides navigational access to related records, but assessment participation remains controlled through separate respondent and approver assignments.
Risk Identification and Review	Enables stakeholders to investigate and act on risks associated with specific AI records.
Policy Monitoring and Violations Review	Allows reviewers to access specific AI records associated with policy violations or compliance investigations.
Governance Decisions and Approvals	Provides record-level access required for stage advancement, approvals, reviews, and governance decisions.

AI Governance-Specific Considerations

AI Governance teams often collaborate across multiple departments and organizations. Many stakeholders require enough context to participate in governance decisions while limiting visibility into sensitive or specialized information.
AI Governance distinguishes between broadly readable information and more sensitive information.

Standard attributes

: Information that helps users understand what a record is, who owns it, and where it is within its lifecycle.

Advanced attributes

Information that is evaluative, risk-related, legally sensitive, or technically nuanced and may require more limited visibility.
ABAC supplements existing RBAC permissions rather than replacing them. Organizations may realize the greatest value from ABAC when module-level roles are intentionally rescoped and ABAC is used to provide record-level access where needed.
Assessment access operates separately from inventory record access. Access to an AI System, AI Agent, AI Model, Dataset, or Project does not automatically grant a user respondent or approver access to associated assessments. These permissions must be assigned independently.
Projects promoted to AI Systems do not automatically carry ABAC assignments forward. Assignments must be re-established on the new AI System record.
AI Automations-discovered records may include pre-populated governance, observability, guardrail, and evaluation attributes that are frequently reviewed by cross-functional stakeholders and are common candidates for ABAC usage.

Supported Objects, Attributes & System Assignments

This section provides a reference for the objects that support ABAC within this solution and the system assignments that can be used to grant record-level access. Supported objects and assignments can be found by navigating to the Assignments screen in Global Settings. For more information, see Managing Assignments with Attribute-Based Access Control (ABAC).

Supported Objects

The following objects support ABAC within AI Governance and can be configured using supported ABAC capabilities: AI agents, AI systems, Assessments, Datasets, Models, Projects, Vendors.

Supported Attributes

The following attributes support assignment-based access and can be used to automatically grant record-level permissions based on ownership, responsibility, participation, or other user relationships. Attributes are organized by object and divided into standard and advanced attribute categories.


Object	Standard	Advanced
AI agents	AI Agent Version, Agent Created Date, Agent Last Updated Date, Agent Region, Agent Type, Audience, Autonomy Level, Business Owner, Coherence, Created By, Created Date, Creation Source, Description, Evaluation/Monitoring Metrics, Groundedness, Has Automated Reasoning Guardrail, Has Content Filtering Guardrail - Input, Has Content Filtering Guardrail - Output, Has Contextual Grounding Guardrail, Has Denied Topics Guardrail - Input, Has Denied Topics Guardrail - Output, Has Prompt Attack Guardrail, Has Sensitive Information Filters Guardrail - Input, Has Sensitive Information Filters Guardrail - Output, Has Word Filters Guardrail - Input, Has Word Filters Guardrail - Output, Hosting Environment, Instructions, Is Deployed, Last Updated By, Last Updated Date, Memory, Memory Retention, Model Binding, Name, Operational Status, Organization, Relevance, Resources, Stage, Task Adherence, Tool Call Accuracy, Tool Selection, Tools, Workflow	AI Guard - Prompt PII Count, AI Guard - Response PII Count, Agent Alias, Agent ID, Agent Source, Aggregated Inherent Risk Level, Aggregated Inherent Risk Score, Aggregated Risk Level, Aggregated Risk Score, Aggregated Target Risk Level, Aggregated Target Risk Score, Automation ID, Data Source ID, Input Token Count, Invocation Count, Output Token Count, Scope
AI systems	Affected Stakeholders, AI Role, AI System Version, Algorithm Type, Business Owner, Business Value, Created By, Created Date, Description, General Purpose AI, Human Oversight, Intended Purpose, Internal or External Use, Last Updated By, Last Updated Date, Master Record/Master Id, Name, Organization, Placing on the Market Form, Purpose(s), Stage, Status, System Learning, Technical Owner, Workflow	Aggregated Inherent Risk Level, Aggregated Inherent Risk Score, Aggregated Risk Level, Aggregated Risk Score, Aggregated Target Risk Level, Aggregated Target Risk Score, Agentic Compound System, Automatic recording of events (logs), EU AI Act High-Risk AI system, EU AI Act Prohibited AI system, EU AI Act Risk Classification, External ID, Processes Sensitive Data, Reasonably Foreseeable Misuse
Assessments	Comments, Deadline, Description, Name, Organization, Record Admin, Reminder, Respondents, Send Email to, Tags, Workflow	Approval Stage, Approver, Assessment Delete Status, Assessment Result, Creator, Date Completed, Date Created, Date Deleted, Date Submitted, Deleted By, Inherent Risk Level, Inherent Risk Score, Open Info Request Count, Open Issue Count, Open Risk Count, Pending Reviews, Residual Risk Level, Residual Risk Score, Stage, Target Risk Level, Target Risk Score, Template Version, Updated Date
Datasets	Annotation, Created By, Created Date, Dataset Version, Description, Function, Last Updated By, Last Updated Date, License, Name, Organization, Owner, Source, Type	Aggregated Inherent Risk Level, Aggregated Inherent Risk Score, Aggregated Risk Level, Aggregated Risk Score, Aggregated Target Risk Level, Aggregated Target Risk Score, Automation ID
Models	Autonomy Level, Biases, Created By, Created Date, Creation Source, Description, Foundation Model, Framework, Generative AI, Hosting Provider, Internal or 3rd Party, Is Master, Last Updated By, Last Updated Date, Limitations, Model Source, Model Version, Name, Organization, Owner, Programming Language, Provider, Reference Link, Stage, Task Type, Type of Learning, Workflow	Aggregated Inherent Risk Level, Aggregated Inherent Risk Score, Aggregated Risk Level, Aggregated Risk Score, Aggregated Target Risk Level, Aggregated Target Risk Score, Automation ID, Coherence, Completeness, Correctness, Demographic Parity Difference, Demographic Parity Ratio, Equalized Odds Difference, Equalized Odds Ratio, Faithfulness, Groundedness, Harmfulness, Has Content Filtering Guardrail - Input, Has Content Filtering Guardrail - Output, Has Contextual Grounding Guardrail, Has Prompt Attack Guardrail, Has Protected Material Code Guardrail, Has Protected Material Text Guardrail, Has Sensitive Information Filters Guardrail - Input, Has Sensitive Information Filters Guardrail - Output, Has Word Filters Guardrail - Input, Has Word Filters Guardrail - Output, Helpfulness, Hub Reference ID, Input Token Count, Invocation Count, Is Entity Attribute Configuration Enabled, Key Hash, Linked AI Systems Count, Master ID, Model ID, Output Token Count, PII Detected Count, Readability, Refusal, Relevance, Response Completeness, Scope, Stereotyping
Projects	Created By, Created Date, Description, Last Updated By, Last Updated Date, Name, Organization, Project Owner, Stage, Status, Time to Review, Type	All system and custom attributes.
Vendors	Business Owner, Created By, Created Date, Last Updated By, Last Updated Date, Name, Organization, Primary Vendor Contact, Status, Type, Vendor Workflow, Vendor Workflow Stage	All system and custom attributes.

System Assignments

The following system assignments are available out of the box and can be used to grant access based on common ownership, collaboration, review, approval, and access-management responsibilities.


System Assignment	Assignment Description	Access Options	Applicable Objects
Assessment approver	Allows user to approve or reject an assessment.	User attributes	Assessments
Assessment creator	Creator of the assessment.	User attributes	Assessments
Assessment respondent	Allows user to respond to the assessments.	User attributes	Assessments
Business Owner	Permits user to edit, delete, and reassign a record.	User attributes	Vendors, AI systems
Collaborator	Allows the user to view the assessment and leave a comment on it.	Sharing	Assessments
Editor	Allows the user to edit and view a record.	Sharing	Vendors, AI agents, AI systems, Datasets, Models
Owner	Permits user to view, edit and reassign a record.	User attributes	AI agents, Datasets, Models
Project editor	Grants permission to edit and view a project record.	Sharing	Projects
Project owner	Permits user to edit, delete, and reassign a project record.	User attributes	Projects
Project record admin	Allows user to edit, delete, and reassign a project record.	Sharing	Projects
Project viewer	Gives ability to view a project record.	Sharing	Projects
Record Admin	Allows the user to edit, delete, and reassign a record. For more information, see Managing Record Admins & Ownership with Attribute-Based Access Control (ABAC).	Sharing	Vendors, Assessments, AI agents, AI systems, Datasets, Models
Risk owner	A risk owner manages a specific risk and implements treatment plans.	User attributes	Assessments
Stage Approver	[Object] stage approver	Workflow	Projects, AI agents, AI systems, Models
Technical owner	Permits user to view, edit and reassign a record.	User attributes	AI systems
Viewer	Allows the user to view a record.	Sharing	Vendors, Assessments, AI agents, AI systems, Datasets, Models

Additional Resources

For a complete list of ABAC resources, see the Attribute-Based Access Control (ABAC) User Guide, which includes foundational concepts, implementation guidance, and procedural configuration articles.

To learn how ABAC is implemented across other OneTrust solutions, explore the following solution-specific articles:

Settings & Administration