Traditional identity and access management is designed to establish who a principal is and what permissions that principal holds. Autonomous AI creates an additional requirement: determining whether a specific action produced during a delegated task remains within the intended authority.
A March 2026 working paper, published in April, proposes an identity-aware governance framework for autonomous agents. It addresses concerns including inherited permissions, independent policy enforcement, multi-step authorization, and verification of actions without relying exclusively on an agent's own account of what happened.
These concerns are especially relevant for enterprise software engineering. A coding agent might be allowed to inspect a repository, create a branch, or prepare a pull request, without being authorized to modify protected workflows or initiate sensitive operations.
A valid identity or token establishes technical access. It does not necessarily capture the full business delegation, the limits of a task, or restrictions that change over time.
Effective authority enforcement therefore needs trusted actor context, a well-defined action, a target resource, applicable policy, and verifiable evidence of the resulting decision. It must also account for changes in authority between the time an agent starts working and the time a consequential operation executes.
Authrizd's GitHub-first approach is an opportunity to implement and demonstrate these principles against real developer workflows.
As enterprise AI transitions from answering questions to modifying systems, the distinction between possessing access and holding authority will become a defining security requirement.
Source: Ramachandran and Mishra, SSRN working paper, posted April 8, 2026
https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6439998
Sources
NEXT STEP