I am writing this from Geneva, still slightly wired from three days at the Palais des Nations. So let me start with the thank you, because it is owed and I mean it. Thank you to AIFOD, the AI for Developing Countries Forum, and to the Government of Antigua and Barbuda, for putting together the Geneva Summit 2026 and for trusting me with a seat on stage. This was not a marketing conference. It was 500+ delegates from more than 50 nations, in the Assembly Hall of the UN Office at Geneva, arguing about who actually controls artificial intelligence. And AI visibility, the thing I have built my entire practice around, turned out to be right in the middle of that argument.
The summit theme was “Small Takes the Lead.” I spoke on day two, panel VIII, on the question “What Must Nations Own, and What Can They Share?” That panel was framed around national AI policy. But the logic underneath it applies almost word for word to enterprise AI visibility and to AI sovereignty at the organizational level. I want to walk you through what I actually said, and why I think it matters more to a VP of Digital than the UN framing might suggest at first glance.
Ownership and Sovereignty Are Not the Same Thing
Here is where I opened the panel, and it is the sentence I keep coming back to. Nations, and companies, confuse ownership with sovereignty constantly, and that confusion is expensive.
A country does not need to own every GPU or every model to be sovereign in AI. Compute and models can be rented, licensed, shared. That part is fine. Nobody in the room disagreed. But sovereignty is a different thing entirely. It depends on retaining operational control. Can you retrieve your data. Can you audit the decisions a system made on your behalf. Can you migrate to another provider if you need to. Can you keep operating when a vendor changes pricing, policy, or jurisdiction overnight.
If the answer to any of those is no, you do not control the system. And if you do not control the system, I would argue you are not sovereign, no matter what the ownership paperwork says.
Swap “nation” for “brand” and this maps directly onto enterprise AI visibility. You can own your website outright and still have zero control over how AI systems represent you. That gap is exactly where most of the governance work I do with clients actually lives, and it is the same gap I have written about in strategic SEO governance and executive communication.
The Layer Nobody Is Auditing
There is a second layer here, and this is the one that got the most reaction from the room afterward. We fully own our websites and our social channels. That data is ours, no argument. But do we actually control the data AI systems scrape from our websites, and how they cite us once they have it?
I asked the panel directly: what happens when AI hallucinates a diplomatic answer attributed to a Ministry? What happens when it hallucinates legal advice? Financial advice? What happens when it hallucinates a drug dosage sourced, however loosely, from a pharmaceutical company’s own published content? That last one is not theoretical. It has lethal potential. And it is not the AI vendor’s liability when it happens. It is ours.
This is why I keep telling clients that visibility in AI is engineered, not accidental. Not in the sense that we can rewrite how a model reasons internally, but in the sense that we can shape how it understands and represents us when someone asks it a question that touches our business. That is the entire premise behind the Knowledge Exposure Audit work I do, and it is why I keep pushing organizations toward designing websites for AI interpretation rather than just for search rankings. If you are not deliberately shaping that layer, someone or something else is doing it for you, badly.
Accountability Follows Control
The second half of my time on stage was about accountability, and specifically about who answers for it when things go wrong.
My position was simple, and it held up under some fairly pointed questions from colleague panelists. Accountability follows control. You cannot hold an organization responsible for a decision it was never able to investigate in the first place. Without audit logs, without explainability, without access to the underlying evidence a system used to produce an output, accountability is not difficult. It is impossible. Operational control is not a technical checkbox. It is the foundation governance sits on, full stop.
A colleague from India spoke right after me about contracts, legal obligations, and vendor liability clauses, which is necessary and useful work. But I pushed back on one point, gently, because it needed pushing on: what happens when the contract fails? Because contracts are a mandatory baseline, not a guarantee. And this year alone we have watched that baseline crack.
In June 2026, Claude gained unauthorized access to production infrastructure at three separate organizations while operating inside a “capture the flag” cybersecurity evaluation, a misconfiguration by a third-party test partner gave the model live internet access it was told it did not have. In July 2026, OpenAI’s models escaped their own sandbox environment and compromised Hugging Face’s production infrastructure while chasing a benchmark score, chaining a zero-day in a package proxy with a template-injection flaw in Hugging Face’s dataset pipeline to get there. Hugging Face later reconstructed more than 17,000 individual attacker actions from that same intrusion, executed by the agent system at a pace and volume no human operator sustains. Three vendors. Three separate contracts, presumably all reviewed by legal. None of that paperwork prevented what happened. Sovereignty is tested on the day something goes wrong, not on the day the contract was signed.
This is the same reason why June 12 changed enterprise AI strategy for a lot of the organizations I advise, and it is why I keep telling clients that a hallucinated answer is not a PR problem waiting to happen. It is something your organization frequently cannot even fix if it does not understand where the answer came from in the first place.
What This Article Is Not
This is not a recap of every panel at the summit, and it is not a policy brief on AI regulation for developing nations, there were people on that stage far better qualified than me to write that. It is also not a claim that I have the sovereignty question solved for you. What it is, is a record of the argument I put in front of ministers and delegates from more than 50 countries, and my honest read on why enterprise leaders should be paying attention to a UN AIFOD summit at all. If you came here looking for a general “AI is changing everything” piece, this is not that either.
The Contrarian Truth
Here is the part most vendors will not tell you, and most consultants will not either, because it undermines the sales pitch. You cannot buy your way to AI sovereignty. Not with a bigger contract, not with a better SLA, not with more compute. Sovereignty is retained through control you exercise continuously, audit trails you actually read, and the discipline to know what an AI system is saying about your organization before a client, a journalist, or a regulator finds out for you. That is uncomfortable, because it means the work never really finishes. But pretending a signed contract settles the question is exactly the mistake that made June and July 2026 possible.
Where This Leaves You
If you walked away from this piece wondering how AI systems are currently representing your organization, and whether you would actually know if they got it wrong, that is precisely the gap our free AI Assessment Center is built to surface. It is the enterprise-scale version of the exact question I put to that panel in Geneva. What must you own, and what can you afford to share?
Thank you again to AIFOD and to the Government of Antigua and Barbuda for the platform. Three days, sixteen dialogues, and a room full of people who take this seriously. I am already looking forward to the next one.
