Policy, standards & oversight / Research brief
NIST proposes identity and access controls for AI agents
NIST researchers proposed giving each software agent a verifiable identity, tying its permissions to a specific task and allowing access to expire or be revoked. They also described when an agent should hand a decision to a person. The authors presented the principles for discussion.
- Federal agency
- National Institute of Standards and Technology
- Source date
- October 27, 2026
- Stage
- Proposed design principles
- Mission
- Cybersecurity
- Audience or eligibility
- Federal identity, security and application owners
Primary source · NIST Cybersecurity Insights ↗
Lighthouse analysis
Conversation starter
Can the identity team trace a delegated action to its sponsor and revoke that authority while the agent is running?
Potential offering
An access-control review could test whether agent permissions expire on time and whether operators can reconstruct who authorized a sensitive action.