The United Nations’ digital agency has decided that AI agents are moving faster than anyone’s ability to trust them, and it wants to do something about it. On 9 July, the International Telecommunication Union used its AI for Good Summit in Geneva to launch an initiative aimed at keeping increasingly autonomous AI systems identifiable, accountable and under meaningful human control.
The mechanism is a focus group, the ITU’s standard vehicle for gathering experts before any formal standard exists. Its job will be to develop frameworks for AI agents that can be recognised for what they are, trusted to act within limits, and kept answerable to the people they act for.
The concern driving it is specific rather than abstract. AI agents are designed to act independently on a user’s behalf, handling everything from scheduling to complex business processes, which means they can also impersonate people and take decisions no one explicitly authorised. As agents proliferate, the ITU argues, the absence of shared rules for identifying and constraining them becomes a systemic risk.
That worry is not unique to Geneva. The wider industry has spent the past year discovering that agent security is largely unsolved, with many organisations unable to say how many agents they are running or what each one is permitted to do.
The 💜 of EU tech
The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!
The ITU is proposing to tackle at the level of international standards what companies have so far been improvising in-house.
The group will pull together technical, policy and legal experts, reflecting the ITU’s preference for multi-stakeholder consensus over top-down mandates.
The stated aim is to bridge a fragmented regulatory landscape in which different jurisdictions are writing incompatible rules for the same technology, often at the same time.
That fragmentation is real. China has already issued national standards covering how agents identify themselves and interact, part of a broader tightening of its agent rules, while other governments and blocs have taken different approaches. A UN body attempting to harmonise the field is, at minimum, an acknowledgement that the current patchwork does not scale.
The work will happen on a deliberate timetable. The focus group is scheduled to hold its first meeting in Paris in November and a second in Geneva in January, with the output likely to be technical recommendations rather than binding regulation.
The ITU sets standards; it does not enforce them, and its influence depends on whether industry and governments choose to adopt what it produces.
That is the built-in limitation of the exercise. A focus group is a slow instrument, and AI agents are a fast-moving target. Between November and whatever the group eventually recommends, the technology will have advanced, and the companies deploying it will have set de facto standards of their own, much as they have with agent identity.
Still, standardisation is where interoperability tends to come from, and identity is exactly the sort of problem that benefits from a shared technical language. If agents from different vendors are going to transact with one another and with humans, some common way of proving what an agent is and who it answers to has to exist. The ITU is betting it can help write that grammar before the market locks in something worse.
The ITU is a natural venue for the attempt, if an imperfect one. It is the same body that spent decades hammering out the standards behind telephone networks, radio spectrum and the internet’s plumbing, the unglamorous agreements that let systems from different countries talk to one another. Applying that machinery to AI agents is a logical extension, even if the pace of the technology strains it.
For now, the initiative is an opening move. No standard has been published, no company has committed to anything, and the first meeting is still months away.
What the ITU has done is plant a flag on a problem the industry keeps deferring, and invite everyone to the table before the autonomous systems start setting the terms themselves.
