Standard
AI Access and Agent Permissions Standard
This standard defines the access governance for AI systems and agents: who can use which AI capability, what an AI agent is permitted to do on the user's behalf, how identity propagates through agentic actions, and how access is reviewed. It extends the identity and access management policy to the AI-specific surface where the actor is not a person but a model invoking tools.
- Type
- Standard
- Version
- 0.0.19
- Date
- 2026-10-27
- Owner
- AI Security Maintainer
- Approving authority
- Governance Library Maintainer
- Classification
- Public
- Confidentiality
- Public
- Category
- AI Governance
- Review frequency
- Annual and upon material agent capability, threat-pattern, or supplier integration change
- Licence
- CC BY-SA 4.0
Read the full text on GitHub (opens in a new tab)
Outline
- 1. Purpose
- 2. Scope
- 3. Principles
- 4. Human access to AI capabilities
- 5. Service-to-AI access
- 6. AI-to-tool access (agentic systems)
- 7. AI-to-data access (retrieval and context)
- 8. AI-to-AI access
- 9. Access review
- 10. Incident-time controls
- 11. Framework alignment
- 12. Limitations
Frameworks it maps to
- ISO/IEC 42001:2023
- EU AI Act
- NIST AI RMF 1.0
- OWASP LLM Top 10
- MITRE ATLAS
- NIST SP 800-53 Rev. 5
- ISO/IEC 27001:2022
- ETSI EN 304 223 V2.1.1 (2025-12)
- Singapore IMDA Model AI Governance Framework for Agentic AI (v1.5, 2026)
- CSA AICM v1.1.0
- CSA CCM v4.1
- Canada TBS Guide on the Use of Agentic AI (comparator)
- OWASP AI Exchange
- SANS CAISG v1.4
Related documents
- AI Security and Risk Standardai
- AI and Agentic Development Security Standardai
- Model Context Protocol Server Registerai
- Model Registryai
- AI Incident Response Planai
- Identity and Access Management Policysecurity
- Privileged Access Management Standardsecurity
- Access Control Proceduresecurity
- Logging and Monitoring Standardsecurity
- AI Human-Oversight Standardai
Next
All Documents