I went to Geneva expecting policy talk. I came back with something closer to a mirror.
Three days in the Assembly Hall at the UN Office at Geneva organized by AIFOD, and the same handful of phrases kept surfacing, session after session, delegation after delegation. Not from one speaker only, but from the room itself too. And the strange part is how directly they mapped onto conversations I have with enterprise clients every week, just with the word “nation” swapped for “brand.”
Here are the three that stuck.
1. Dependency, not resources, is the real vulnerability
I heard the same version of this constantly through the conversations and panels. We must end dependency. How much more sovereignty are we ready to give away. Our biggest problem is lack of control. We don’t want to be overtaken by big players.
None of that was really about GPUs or compute budgets. It was about the slow, almost invisible transfer of control that happens when you outsource the thinking layer and not just the infrastructure layer. A nation can have a functioning AI strategy and still be entirely dependent on someone else’s model, someone else’s training data, someone else’s definition of what counts as accurate.
Enterprises do the exact same thing without noticing. You publish content, you build a site, you assume the story stops there. It does not. The dependency shows up later, when an AI system answers a question about your company using an outdated page, or worse, using someone else’s characterization of you instead of your own. You do not lose control all at once. You lose it one unindexed, unaudited page at a time, by intentionally giving up control you should keep firm.
2. Governance beats participation, every time
Second phrase that kept coming back: we must have governance, not be spectators all the time. And close behind it: we don’t want to allow big organizations to tell us what we should think, do and feel.
That are harder statements than it sounds. Being present in a system is not the same as governing your place in it. Plenty of nations in that room have AI policy documents. Fewer have actual mechanisms to audit what is happening to their data, their citizens, their information once it enters someone else’s model. Presence without governance is spectating with extra steps.
I see this constantly on the enterprise side too. A company can rank, get cited, show up in AI answers, and still have zero governance over how it is being represented. Showing up is not the same as owning the outcome.
3. You do not need to own the machine. You need to own the information layer.
This was the core of what I argued on stage, and Geneva only sharpened it for me. Small nations and enterprises do not need to own every GPU, every LLM, every layer of infrastructure. That is the race most of them can’t win, and chasing it wastes energy you can better spend somewhere else.
What is winnable, and what actually matters, is the information layer. This is the exact dimension in which big and small are equal. And I can argue that smaller ones are even in much better position because of their agility. Every nation in that room has public channels. Every enterprise has a website. That is sovereign territory already, whether anyone treats it that way or not. The problem is not the absence of ownership. It is the absence of control over how that owned material gets understood and represented once an AI system reads it.
And this is where it stops being theoretical and starts being liability. When an AI system hallucinates a diplomatic position that did not come from the ministry that owns it, that is not a minor hallucination. When it invents legal guidance attributed to a law firm, or financial advice attributed to a bank or financial institution, or a drug dosage attributed to a pharmaceutical company, that is not a technical glitch anyone can shrug off. That is responsibility, and in some contexts, potentially real legal liability. Technology should supplement human judgment here, not substitute for it, and AI should not be making decisions or forming positions that compete with the humans who are actually accountable for them.
Owning your information layer, presenting it clearly, keeping it accurate and current, is not an SEO requirement anymore. It is the difference between an AI system representing you correctly and an AI system inventing something you now have to publish correctly, or defend.
What I am taking back to client work
This is exactly the gap I built the AI Visibility Inspector and Knowledge Exposure Audit to close, and it is also why I finished the Free AI Assessment Center before I left for Geneva, in case someone in that room wanted to see what it looks like in practice. It is live now and available for everyone wishing to check them.
The nations in that Assembly Hall were arguing about sovereignty at a geopolitical scale. The argument underneath it, the one that applies to every organization with a website, is smaller and more immediate. You already own the information. The only open question is whether you are actually governing it, or just hoping it gets represented correctly. And this difference is huge.
FAQ
Dependency on external AI infrastructure, the difference between participation and governance, and control over how organizations and nations are represented by AI systems.
It means owning and actively governing the accuracy of your public information (websites, published content, official channels) rather than owning the AI infrastructure itself. Most organizations already have this layer; few actively manage how AI systems interpret it.
When an AI system generates incorrect diplomatic, legal, financial, or medical information and attributes it to a real organization, that organization can face real-world consequences and responsibility for the error, regardless of whether they generated the content themselves.
