In early July, over four days, global AI governance was written in Geneva. The UN’s first global Dialogue on the subject was held at Palexpo on 6 and 7 July. The ITU’s AI for Good summit followed until the 10th, with some 12,000 participants from 170 countries. António Guterres summed up the stakes in a phrase that travelled across the media: govern AI by design, or drift by default.
Meanwhile, fifteen minutes away by tram, hundreds of Geneva SMEs were asking a distinctly more down-to-earth question. Does any of this concern them? Or is it another of those international gatherings the city specialises in, which change nothing in their daily lives?
The honest answer comes in two parts. No, nothing changes in how you operate tomorrow morning. But two very real deadlines already concern you, and the noise of the summit is drowning them out.
Global AI governance: what was decided at Palexpo
Let us start again from the beginning, because the landscape has grown dense fast. In August 2025, the UN General Assembly created two mechanisms by consensus. On one side, an international scientific panel of 40 members, somewhat on the model of the IPCC for the climate. On the other, the much-discussed Global Dialogue that has just met for the first time.
The scientific group is co-chaired by Yoshua Bengio, among others. It delivered its first report on 1 July, and its conclusion deserves to be read as it stands: AI capabilities are advancing faster than scientific understanding, and faster than governments’ ability to adapt.
So much for the global picture. It is real, it is important, and it is playing out literally on our doorstep. The Confederation will also host a global AI summit in 2027, with the support of the canton of Geneva.
Principles in Geneva, rules elsewhere
Except that a UN dialogue produces principles, not obligations. No discussion at Palexpo creates a rule that applies to a fiduciary firm in Plainpalais or to a mechanical workshop in Yverdon. What creates rules is the law. And the law moves on an entirely different timetable.
On the Swiss side, the Federal Council set out its position in February 2025. It ratifies the Council of Europe Convention on AI, signed in March 2025. It adapts the laws sector by sector. And it relies on non-binding measures such as codes of conduct. So no sweeping horizontal law on the European model, for the time being.
The federal departments have until the end of 2026 to submit their work. A Swiss legal framework specific to AI will therefore not exist before 2027 at the earliest. Switzerland regulates in small touches, in the country's tradition. As for the Council of Europe Convention, it targets state uses of AI first, not yours.
The two texts that already concern you
Does that mean you can use AI with no framework at all until then? No. And that is the first point many executives miss.
The Federal Act on Data Protection already applies to any processing of personal data by an AI tool. If your assistant copies client files into ChatGPT to draft letters, you are within the scope of the LPD. Add to that the questions around transferring data abroad. This subject is not waiting for an AI law: it has existed since September 2023.
The second blind spot concerns SMEs that sell in Europe. The European AI Act is a horizontal law, and it applies extraterritorially. A Swiss company that supplies an AI system used in the Union falls under its obligations, with no European headquarters. The same goes for a product sold there with an embedded AI layer.
The Federal Council is taking this into account, as it happens. Its roadmap explicitly mentions compatibility with trading partners, to avoid isolating Swiss companies. If your market is French-speaking Switzerland, the subject remains theoretical. If you export software or machines with embedded AI, it is not theoretical at all. In that case, the AI Act deserves an hour of serious reading, a real one. And if you would rather we looked at it with you, that is the ground covered by our AI and automation support.
The opportunity nobody is looking at
One question remains that nobody asks from this angle. What does the Geneva concentration change as an opportunity, rather than as a regulatory threat?
The canton was not mistaken about this: its delegation at Palexpo presented the local ecosystem to visitors. For an SME in the region, that institutional density has very concrete consequences. International organisations and NGOs buy services, including digital ones. And they are happy to buy them from local providers who understand their confidentiality and neutrality constraints.
A summit of 12,000 people also means needs in translation, events, development and communication. Not counting the implicit label. Being a tech company from the region hosting global AI governance makes for a very good story with a client in Zurich or Lyon.
In practical terms, where does this institutional clientele go? Most purchasing across the UN system passes through UNGM, the United Nations Global Marketplace. Registration is free and within reach of an SME, provided you take the time to document your references properly. The large Geneva NGOs publish their tenders separately, often on their own sites. And the canton’s economic development office supports local companies wanting to approach this market. None of this calls for a lobbyist. It calls for a clean file and consistency.
There is one more diffuse but well-documented effect: talent. The people who come to Geneva to work on these subjects form a pool that local SMEs can reach. Not by competing with Google salaries, but by playing the card of purpose and proximity. You see it at every major conference: some of these profiles then look to stay in the region, and turn towards local companies.
Three moves to make before December
So what should you do with all this? Three reasonable moves, with no consultant and no budget.
Start by taking stock of how AI is actually used, including the unofficial use by your employees. That is where the LPD risk sits, not in your official strategy.
Then check whether one of your products or services touches the European market with an AI component. If so, the AI Act moves ahead of everything else, and it is worth dealing with this year.
And if you sell to international organisations, or want to, register now in the local purchasing channels. The 2027 summit is being prepared in 2026. Applications filed early are the ones that later show up in supplier lists when budgets are released. And those lists fill up long before the official announcements.
The rest, the declarations, the panels, the resolutions, all of that will run its course with or without you. Global AI governance is being written in Geneva, and that is a perfectly legitimate source of local pride. What matters for your company is telling apart what is decided there from what applies to you. On that second front, the two texts that count already exist. The LPD (Swiss Data Protection Act) here, the AI Act at our neighbours’.
