The United Nations' International Telecommunication Union staged its 10th annual AI for Good summit in Geneva this week, filling a 106,000-square-meter convention center with humanoid robots, Tesla Cybertrucks, UN rescue helicopters, and a 44-member commission tasked with steering global AI policy. The event's scale underscored a widening gap between where the technology is heading and where multilateral consensus can follow. While Silicon Valley executives testify in Washington about superintelligence and the White House shuffles chip export controls, the ITU is trying to reframe AI as a development tool for the roughly four billion people outside the US-China-Europe axis.
ITU secretary-general Doreen Bogdan-Martin opened with the summit's founding thesis, that responsibly deployed AI could address hunger, disease, and climate change. A decade in, that thesis is under pressure from the technology itself, and from the concentration of the companies building it.
The new 44-member commission, cochaired by Rwandan president Paul Kagame and Salesforce CEO Marc Benioff, is the summit's headline deliverable. It is meant to translate the AI for Good brand into something with a governance function rather than a conference function. What the commission will actually enforce, and against whom, was left largely undefined across three days of panels.
Key facts
- 01The UN's AI for Good summit hit its 10th year, spanning a 106,000-square-meter convention center outside Geneva.
- 02The ITU announced a 44-member commission cochaired by Rwandan president Paul Kagame and Salesforce CEO Marc Benioff.
- 03Amazon CTO Werner Vogels was interrupted by pro-Palestine protesters who stormed the keynote stage before being removed.
- 04Speakers from Access Now, the World Bank, and IEEE pushed back against 'governance theater' and opaque public-sector tech deals.
- 05Tesla Cybertrucks, UN rescue helicopters, and running humanoid robots shared the convention floor.
Criticism came from inside the tent. Giulio Coppi, senior humanitarian officer at Access Now, argued on the sidelines that humanitarian organizations and public agencies have grown too dependent on major cloud vendors, calling for an end to treating tech companies as friends. He pointed to a decade of opaque, multimillion-dollar deals funded by public money and said buyers often cannot describe what is inside the stacks they have licensed.
The most visible dissent came during a keynote by Amazon chief technology officer Werner Vogels, when pro-Palestine activists stormed the stage alleging Amazon technology is being used by Israel against Palestinians. Security removed them. The interruption highlighted how enterprise cloud contracts, once treated as neutral infrastructure decisions, have become political flashpoints for AI vendors selling into government.
Harvard University engineering professor Vijay Janapa Reddi pushed on a different failure mode: the gap between AI demos and AI systems that hold up in production. He argued that 'good' is not an engineering specification, and that a plane that flies for five minutes is not a plane. The point echoed across sessions on hardware and access, where speakers argued compute is now a development-infrastructure problem, not a technology problem.
Syed Munir Khasru, chairman of the Institute for Policy, Advocacy, and Governance, said compute for all has to be treated as development infrastructure or the phrase means nothing. Others noted that most large language models remain structured around English, making smaller local models on cheaper hardware essential if AI is to serve communities beyond the richest markets. That framing put the summit's politics squarely on chips, standards, and who gets to shape both.
Gilles Thonet, deputy secretary-general of the International Electrotechnical Commission, made a case for embedding human rights into technical standards work rather than treating them as a separate policy track. Anja Kaspersen of IEEE argued the most consequential AI decisions are made inside technical standards and procurement choices, not UN assemblies, and proposed what she called middleware — a connective layer translating high-level principles into verifiable enforcement.
Jeremy Ng, counsel for AI and the digital economy at the World Bank, was blunter, saying AI impact assessments have to become practical tools with real teeth rather than governance theater or a box-ticking exercise for tech giants. That framing lines up with the direction regulators in Brussels and several capitals are taking, where paper commitments are being pushed toward audit-grade requirements.
Access-and-export politics ran through the corridors. The Trump administration has implemented and then lifted export controls on frontier models, and China is reportedly considering making its open-weight releases less open. Tightening those flows leaves smaller economies dependent on foreign platforms, foreign standards, and foreign pricing.
Bogdan-Martin closed her opening with a line that captured both the summit's ambition and its limits, saying no single stakeholder can shape AI alone and that the effort needs builders. The builders, however, were mostly not at the podium. They were shipping.
The image that lingered from Geneva was not the 44-member commission or the middleware pitch. It was the humanoid robots running between booths on the convention floor while diplomats debated definitions upstairs. The commercial layer of AI is compounding faster than any multilateral body can convene, and the UN's bet is that a coalition of governments, standards bodies, and one large enterprise software CEO can build enforcement infrastructure quickly enough to matter. That bet gets harder every quarter the frontier labs ship, and the summit's value now depends on whether the commission produces auditable requirements or another round of principles by the time it meets again.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
