OpenID Connect for Agents
OIDC-A
A proposal (an arXiv paper, not an OpenID Foundation specification) to extend OpenID Connect with claims, capability metadata and discovery for LLM-based agents.
In our briefings
The claim each of the newest briefings on OpenID Connect for Agents would have you leave with.
The OpenID Foundation's own community group draws the line on what today's identity standards do for AI agents: they 'can work for synchronous agents using multiple tools across a single trust domain, but not asynchronous or multi-domain contexts.' In plain terms, OAuth 2.1 handles an agent that works inside one organization while its user is present. It falls short when the agent runs unattended, crosses into another organization's systems, or acts for several people at once. Everything vendors now sell as agent identity for the simple case rests on standards that already existed; the hard cases have drafts and proposals.
OAuth handles an agent in one company, working while you watch. The OpenID paper says that is where it stops.BI-FIND-2025-0054
MacGuire reports that xChk’s production OpenID Connect (OIDC) login sends human identity portfolios to relying parties. Agents authenticate with per-agent API keys and use the authorize/attest API for runtime governance. Attestation and delegation-chain embeds use a separate first-party token-refresh surface; full non-human entity types in relying-party OIDC tokens remain future work. An integrator cannot infer agent OIDC support from the paper’s combined token illustration.
xChk ships human OIDC portfolios and a separate agent approval APIBI-FIND-2026-1637
Yu, Geng and Knottenbelt propose that an AI agent should never hold a reusable secret, only the right to cause one approved use of it. In their Secret-Use Delegation Protocol (SUDP), the agent proposes an exact description of the operation (for a GitHub call: method, repository, endpoint, branch and a hash of the body). The user signs a hash of that description, with a fresh challenge, on a WebAuthn authenticator. A separate custodian holding the sealed secret checks the grant, runs the call once and hands back only the result, so a captured grant is "redeemable only against the exact operation it was signed for." The price is a user gesture for every action, which the authors accept because the protocol is aimed at high-authority secrets whose misuse already warrants a person's oversight per action. It does not stop a compromised agent from proposing a harmful operation that the user then approves as displayed, and any field left out of the description is outside the guarantee.
An Imperial and Oxford protocol would give AI agents one approved use of a secret, never the secret itselfBI-FIND-2026-1257
Cited in 4 briefings
Every briefing that discusses this term, newest first.
- OAuth handles an agent in one company, working while you watch. The OpenID paper says that is where it stops.
Article · · OpenID Foundation · Tobin South et al. · Oct 2025
- xChk ships human OIDC portfolios and a separate agent approval API
Paper · · arXiv · Sean MacGuire · Jul 15, 2026
- An Imperial and Oxford protocol would give AI agents one approved use of a secret, never the secret itself
Paper · · arXiv · Xiaohang Yu et al. · Apr 27, 2026
- One author's OpenID Connect for Agents proposal checks each delegation hop but makes the chain optional
Paper · · arXiv · Subramanya Nagabhushanaradhya · Sep 30, 2025