AI Agent Identity

Summary: The emerging concept of giving AI agents their own verifiable digital identity — separate from, but bound to, the human, company, or institution they act for — so their actions can be scoped, authorized, and audited.

Sources: 2026-09-04-ai-agent-identity-news.md, 2026-09-07-ai-agent-identity-news.md, 2026-09-08-ai-agent-identity-news.md

Last updated: 2026-09-08

As AI agents increasingly act on behalf of people and organizations (browsing, filling forms, preparing documents, initiating payments), several governments and standards bodies are converging on the idea that agents need a first-class identity distinct from shared credentials or a human’s own login.

Core elements described across sources

A unique identifier or credential for the agent itself, rather than the agent silently reusing a human’s password or API key (source: 2026-09-04-ai-agent-identity-news.md).
Binding to a responsible party — the identity is tied back to the person, company, or institution that authorized the agent, so actions remain traceable (source: 2026-09-04-ai-agent-identity-news.md).
Scoped, limited permissions rather than full delegation of a user’s rights — see agent-authorization-and-delegation (source: 2026-09-04-ai-agent-identity-news.md).
Auditability — actions taken by the agent should be logged and attributable (source: 2026-09-04-ai-agent-identity-news.md).

Estonia’s public correction sharpened this concept materially. Giving an agent a first-class identity does not have to mean giving it a human-style legal identity: Estonia will not issue personal identification codes to agents, and instead links an agent to the person it represents, with liability staying human (source: 2026-09-08-ai-agent-identity-news.md, citing ERR News). See estonia-ai-agent-id.

A related distinction from the standards side: discovery, authentication, authorization, and accountability are separate layers, and proposals that appear to disagree sometimes address different ones (source: 2026-09-07-ai-agent-identity-news.md, citing Cloud Security Alliance). See agent-standards-and-interoperability.

Where this is happening

estonia-ai-agent-id — the most concrete national programme; note the 2026-09-08 correction withdrawing the “AI personal identification code” framing.
agent-standards-and-interoperability — the technical layer: OWASP ACS, IETF DAWN, WIMSE, SPIFFE, OAuth delegation.
united-states-ai-agent-policy — NIST guidance pointing to OAuth 2.0 and SPIFFE as existing building blocks.
china-ai-agent-policy — a national standard covering identity management for interconnected agents.
singapore-ai-agent-policy — a sandbox finding that identity/authentication infrastructure isn’t yet ready for agents.
malaysia-ai-agent-policy — framing agents as “non-human identities” in need of legal accountability structures.

Open questions

Partly answered (2026-09-08): on whether proposals have moved beyond the advisory stage — Estonia’s Aruait now has a published scope and named deliverables (Identity 2.0, a trust registry, interoperability standards, a service-to-service pilot), though no deployed system is yet described (source: 2026-09-08-ai-agent-identity-news.md, citing RIA). The status of the other jurisdictions still needs verification.
Still open: how agent identities interoperate across jurisdictions. Estonia lists interoperability standards as a goal, but no source describes cross-border recognition — needs verification.
New (2026-09-08): how a national trust registry relates to workload-identity standards such as SPIFFE is unaddressed in all sources so far.