# Authenticated Delegation and Authorized AI Agents

**Authors:** Tobin South, Samuele Marro, Thomas Hardjono, Robert Mahari, Cedric Deslandes Whitney, Dazza Greenwood, Alan Chan, Alex Pentland  
**arXiv:** [2501.09674](https://arxiv.org/abs/2501.09674)  
**Date:** 2025-01-16  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2501.09674v1/

## Key points

- The rapidly evolving landscape of autonomous AI agents presents challenges in authorization, accountability, and access control in digital spaces.
- New standards are required to establish whom AI agents act on behalf of and guide their usage appropriately for safe and effective deployment.
- A novel framework has been introduced for authenticated, authorized, and auditable delegation of authority to AI agents.
- This framework allows human users to securely delegate and restrict permissions and scope of agents while maintaining clear chains of accountability.
- Building upon existing identification and access management protocols such as OAuth 2.0 and OpenID Connect, this framework extends them with agent-specific credentials and metadata for compatibility with established authentication infrastructure.
- A framework has been proposed for translating flexible natural language permissions into auditable access control configurations to enable robust scoping of AI agent capabilities across various interaction modalities.
- Authenticated delegation integrates existing approaches such as AI agent IDs and credentials with proof-of-personhood for human users and content provenance methods to anchor the actions of AI agents to verifiable human principals.
- The technical framework extends internet-scale authentication protocols like OAuth 2.0 and OpenID Connect to address the unique challenges of authenticating AI agents while maintaining compatibility with existing internet infrastructure.

## Summary

The rapidly evolving landscape of autonomous AI agents presents pressing challenges surrounding authorization, accountability, and access control in digital spaces. To address these issues and ensure the safe and effective deployment of AI agents, new standards are required to establish whom these agents act on behalf of and guide their usage appropriately. This not only safeguards online spaces but also unlocks the potential value of delegating tasks to autonomous agents. A novel framework has been introduced for authenticated, authorized, and auditable delegation of authority to AI agents. This framework allows human users to securely delegate and restrict permissions and scope of agents while maintaining clear chains of accountability. Building upon existing identification and access management protocols such as OAuth 2.0 and OpenID Connect, this framework extends them with agent-specific credentials and metadata. This ensures compatibility with established authentication and web infrastructure. Furthermore, a framework has been proposed for translating flexible natural language permissions into auditable access control configurations. This enables robust scoping of AI agent capabilities across various interaction modalities. By implementing this practical approach, immediate deployment of AI agents is facilitated while addressing crucial security and accountability concerns. The goal is to ensure that agentic AI systems only perform appropriate actions while providing digital service providers with a tool to enable AI agent interactions without risking harm from scalable interaction. The documentation, safety, and governance of agentic AI systems have become paramount in recent research efforts. Early frameworks like datasheets, model cards, and data statements have laid the foundation for understanding bias, privacy concerns, copyright issues in AI systems. However, as AI systems become more agentic, new frameworks are needed to document their evolving capabilities, decision-making processes, and potential risks. To combine these solutions effectively,

authenticated delegation integrates existing approaches such as AI agent IDs and credentials with proof-of-personhood for human users

and content provenance methods. By anchoring the actions of AI agents to verifiable human principals and recognized AI-specific credentials through granular permission sets and clearer accountability chains, the paper introduces a concrete technical framework that extends internet-scale authentication protocols like OAuth 2.0 and OpenID Connect to address the unique challenges of authenticating AI agents while maintaining compatibility with existing internet infrastructure. In conclusion, this practical framework for authenticated delegation to AI agents offers a comprehensive solution to the challenges surrounding authorization, accountability, and identity verification in digital spaces. By extending existing protocols with specific mechanisms for delegating authority from users to AI agents while ensuring clear chains of accountability through token-based authentication framework comprising user ID tokens, agent-ID tokens, and delegation tokens provides a robust foundation for verifying agent identities, controlling permissions, maintaining audit trails, and supporting granular and robust scoping of AI agent capabilities.