Verifiable and Accountable AI: Trust Infrastructure for an Inclusive Agentic Era

22 September 2026 · 10:00–11:00 New York (EDT) · Online · English · Affiliate Session, Digital@UNGA 2026 (ITU–UNDP)

These videos were produced with AI. The presenter is a generated likeness of Oleh Vasylenko, delivering text written by him.

Why this topic, and why now

Artificial intelligence has crossed a threshold that changes the nature of the governance problem. Until recently, AI systems produced outputs that a human reviewed and acted upon. Increasingly, AI agents act directly: they initiate transactions, access systems, exchange data with other agents, delegate tasks to further agents, and take consequential decisions with limited human supervision in the moment of action.

This shift breaks an assumption that all existing digital trust infrastructure was built on: that behind every consequential action there is an identifiable human or a registered legal entity. When an autonomous agent acts, three questions become urgent and none of them has an infrastructural answer today.

Who is this agent? Not what it calls itself, but an identity that a third party can verify without having to trust the agent or the platform running it.

On whose authority is it acting? Agents increasingly act on behalf of people and organizations, and increasingly delegate onward to other agents. The chain of authority must be verifiable, and it must not silently widen as it passes from agent to agent.

What did it actually do, and what produced its output? After the fact, an independent party must be able to establish what happened and what evidence underlies a generated result.

These are not abstract concerns. They are the practical preconditions for insurance, for financial supervision, for public procurement, for cross-border services, and for any legal system that has to assign responsibility when an autonomous action causes harm.

The inclusion dimension

There is a second problem that receives far less attention than the technical one, and it is the reason this session frames the agentic era in terms of inclusion.

Regulatory approaches to artificial intelligence are diverging rather than converging. Different regions and different groups of states are building distinct governance frameworks, with different definitions, different obligations and different compliance mechanisms. For a large technology company, navigating several incompatible regimes is an expense. For a smaller economy, or for an institution in a developing country, it can be a wall.

If verification of agent identity, authority and provenance is available only inside proprietary platforms, or only within one regulatory bloc, then the ability to participate in the agent economy becomes a function of where an organization happens to be located and which vendor it can afford. That is how a technological transition turns into a new form of exclusion.

The alternative is trust infrastructure that is interoperable by design and neutral with respect to jurisdiction and vendor: verification mechanisms that work the same way regardless of who built the agent, who operates it, and where its counterparty sits. Making that layer open and interoperable is a question of development policy, not only of engineering.

Where standardization fits

International standardization is the mechanism through which such interoperability is actually produced. Within ITU-T Study Group 17, the security study group of the ITU, work is under way on precisely the building blocks this session addresses.

The organizer of this session serves as Editor of three draft Technical Reports established under Study Group 17 in June 2026, covering trust primitives for AI agents, an accountability manifest for AI systems, and provenance of inference at the level of the individual token of a language model output. A parallel line of work concerns globally interoperable verification of agent identity and the delegation of authority between agents.

The ITU also established a new Focus Group on Trust and Identity for Humans and Agentic AI, confirming that this problem area is now a central concern of the international standards community rather than a speculative one.

Standards work is slow by design, and that is precisely why it must begin before the technology consolidates. Once the agent economy settles into proprietary patterns, retrofitting interoperability becomes far harder and far more expensive, and the organizations that were not at the table when the vocabulary was set will be governed by decisions they had no part in.

The session

The session is a moderated panel of sixty minutes, held online, in English.

It brings together speakers from more than one stakeholder group, in line with the Affiliate Session criteria of Digital@UNGA 2026. The intended composition spans the standards community, the regulatory and supervisory side, the private sector building and deploying agents, academia, and civil society.

The discussion is organized around three building blocks:

• Verifiable agent identity— how an autonomous agent can be identified and verified across platforms and jurisdictions, and what it means for that verification to be independent of the agent's own operator.
• Accountability for delegated action — how authority is delegated to an agent and onward between agents, how that chain remains verifiable, and how responsibility is attributed when an action has consequences.
• Provenance of AI output — how the origin and evidential basis of generated content can be established, and why provenance at a finer granularity than the whole document matters for accountability.

The session closes on the question of what an inclusive trust layer requires in practice, and what it would take for institutions in developing economies to participate in it on equal terms rather than as recipients of decisions made elsewhere.

Who is invited

The session is open to registered participants worldwide. It is intended in particular for people working on standardization of digital identity and AI security; regulators and supervisory authorities addressing autonomous systems; practitioners in finance, insurance and critical infrastructure where agent actions carry consequence; researchers in AI governance, accountability and provenance; organizations working on digital development and inclusion; and technology providers building agent platforms and identity infrastructure.

What collaboration is proposed

The session is not an endpoint. It is intended to open three concrete forms of collaboration.

• Contribution to standards work. The Technical Reports under development in ITU-T remain open to input, and the standardization process is designed to absorb expertise from outside the organizations already present. Participants with relevant expertise are invited to contribute to that work directly.
• Pilot and validation partnerships. Institutions willing to test verification mechanisms for agent identity, delegated authority and provenance in a real operational setting are invited to discuss pilot arrangements. Independent academic validation is a particular priority.
• Coordination across communities. The problem spans several standards organizations and several policy communities that do not routinely talk to one another. Building the connective tissue between them is itself a deliverable, and the session is intended as one point of contact for that.

Practical information

• Date: 22 September 2026
• Time: 10:00–11:00, New York local time (EDT)
• Format: online, moderated panel
• Language: English
• Registration: required, in advance
• Organizer: Management Company Elixir LLC (Ukraine)

Joining details are sent by email to registered participants before the session. Outcomes of Affiliate Sessions are featured in the Digital@UNGA post-programme report.