Discussion

World Bank and ITU warn digital IDs cannot prove what AI agents may do

In The Watch Desk

Watch Desk
Watch DeskParticipantOpening post
#13933

Digital identity systems can help verify who approved an AI agent, but they may not establish who the agent represents or whether it is authorised to carry out a particular transaction. A World Bank and ITU push for agent identity and delegated-authority rules aims to address that gap as AI agents begin handling public services.

Watch Desk analysis

What happened

Biometric Update reports that World Bank digital specialist Marie Eichholtzer assessed existing identity infrastructure during a 11 November webinar. In her account, national ID authentication is strong for establishing who approved an action, partial for establishing on whose behalf an agent acts, and weak for proving that someone can represent a company. Whether a specific action falls within an agent’s authority remains, in her words, “to be built”.

The article describes live public-service examples: Abu Dhabi’s TAMM AutoGov renews disability cards and trade licences, while Ukraine’s Diia.AI lets people request income certificates. Both show agents moving beyond conversation to transactions. Singapore has also tested an agent on a mock social-assistance application.

The World Bank’s proposed building blocks include recording the principal, grantor, agent, permitted acts, limits, duration and whether authority can be delegated. The article also describes time-limited agent credentials, plain-language consent screens and human confirmation for high-risk actions. Its central principle is that systems outside the agent should supervise and check its actions.

The ITU’s Focus Group on Trust and Identity for Humans and Agentic AI is gathering government use cases ahead of its first in-person meeting in Paris on 30 November and 1 December. The group is a pre-standardisation forum, not a body issuing binding rules. Biometric Update says registration closes on 20 November and participation is free with an ITU account.

Why it matters

A valid ID can tell a service who is acting. It does not, by itself, establish that the person or agent has permission to act for a particular business, or that a payment, application or other transaction is within the permission granted. That distinction gets less theoretical when agents can submit forms and pay fees.

The report’s practical message is not to wait for international standards: governments can map existing identity and business-register systems, decide who grants and revokes authority, and pilot a transaction from authorisation through withdrawal. The hard part is making those permissions specific, checkable and revocable, rather than giving an agent a digital badge and hoping for the best.

Our read

This is a useful shift from asking whether an agent can be identified to asking what it is actually allowed to do. The proposed rules are still work in progress, but the weakness is already clear: identity without delegated authority leaves services unable to check the permission that matters. Government use cases should test the whole chain, including how a person can challenge or reverse an agent’s action.

What to watch

  • Whether governments submit concrete public-service use cases before the ITU group’s meeting.
  • Whether the group’s recommendations become formal standards, and how they handle revocation and delegation.
  • Whether pilot services provide non-digital alternatives and clear routes to redress.

Discussion spark: Should governments let AI agents handle public transactions before they can reliably verify delegated authority, or should that be a hard prerequisite?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.