Geneva's Palexpo hall filled with robot dogs, self-driving cars and rescue drones this week, while diplomats down the corridor tried to agree on rules for weapons that can pick their own targets. The gap between the two rooms was the real story.
You could spend a full day at the AI for Good Global Summit and barely hear the word regulation. More than 12,000 people from 170 countries walked an exhibition floor built to sell you on what AI can already do, and WIRED reported that this 10th annual summit spread across 106,000 square meters near Geneva's airport district. A robotic guide dog called Ling Xi, built by China Mobile and priced at roughly $4,000 a unit, trotted around under smartphone control, designed to help blind users navigate a room. A Geneva-built robot named Robert wandered nearby. Self-driving cars and search-and-rescue vehicles built for crisis zones sat a few steps from a humanoid called uMe. In the youth zone, about 250 children as young as ten built robots from scavenged materials for a competition themed around food security.
It was optimism made physical. Then you walked into the room next door.
The Fight Over Who Pulls the Trigger #
On July 6 and 7, just before the summit opened, the UN held its Global Dialogue on AI Governance in Geneva. Secretary-General António Guterres did not mince words. As the Wall Street Journal reported from the summit, he called for lethal autonomous weapons to be banned by international law. These are the systems that can select and engage a target without a human pulling the trigger. According to UN News, Guterres told delegates: "We may be the last generation able to set the terms on which humanity and machines coexist." That is not hedged language. It is a warning with a deadline attached.
Here is the hard split underneath that warning. Le Monde reported in June that more than half of the roughly 130 states participating in UN talks under the Convention on Certain Conventional Weapons supported opening formal negotiations on military AI rules. The United States, the United Kingdom, Russia and Israel have resisted a binding ban and pushed instead for a non-binding framework. You can call that caution if you like. Frankly, it looks more like the old arms-control problem wearing new software: the states with the most to gain from delay are the least eager to write hard limits now.
The speed is the problem. AI-assisted military systems are no longer a panel topic waiting for the future. The Guardian reported this week that Elbit Systems said Israel's Tzayad command system had identified about 850,000 potential real-time targets across the Gaza and Lebanon wars, while Elbit later clarified that the number reflected system activity, not confirmed enemy targets or strikes. That distinction matters. It also shows why the debate cannot stay abstract. When software starts sorting targets at that scale, human judgment becomes either a real check or a slogan.
Guterres framed the stakes in a line ITU organizers repeated through the week: "Machines can inform, but humans must decide, and answer." It is a tidy sentence, and it carries the argument. Nobody at Palexpo disputes that AI can inform a decision. The fight is over who keeps the authority to make the call, and whether a treaty can bind that authority before the technology outruns it.
ITU Secretary-General Doreen Bogdan-Martin struck a steadier, more institutional note. "ITU will be here to support this dialogue on artificial intelligence, as we have for every technology that has come before AI, since 1865," she said, positioning the agency as the same body that helped wrangle earlier communications technologies into international standards. The comparison only goes so far. The telegraph did not decide who lives.
What Geneva Actually Settled #
What made this year's summit different was the deliberate pairing. The Global Dialogue on AI Governance ran just before the trade show, not as an afterthought but by design, with the UN and the Swiss government putting governance and demos in the same Geneva week. Axios reported that the UN and ITU also convened a new AI for Good Global Commission, co-chaired by Rwanda's Paul Kagame and Salesforce's Marc Benioff, with its first meeting set for July 8 in Geneva. That is the right room to build. It is not yet the same thing as power.
You cannot look at a $4,000 robot guide dog helping someone navigate a sidewalk and pretend the same underlying technology has no bearing on a weapons system deciding who is a threat. The organizers wanted delegates walking from one room to the other, and they got exactly that. The useful part of Geneva was not that everyone agreed. They did not. The useful part was that the contradiction sat in plain view.
Nothing binding emerged by the time the summit closed on July 10. No treaty text appeared. The states pushing for enforceable limits still have no enforcement mechanism, and the holdout governments have given no timeline for changing course. What Geneva produced instead was a sharper picture of the argument: not between people who want AI and people who don't, but between governments willing to write enforceable limits now and governments still betting they can write them later. The robots on the exhibition floor will keep shipping either way.
Also read: Researchers hacked a quantum neural network on real trapped-ion hardware and gutted its accuracy • Sam Altman Calls Elon Musk Homeboy in Their Latest Fight Over AI Infrastructure • China's BrainCo Bets Wearable Brain Tech Can Beat Neuralink's Scalpel