The United Nations’ International Telecommunication Union used its 10th AI for Good summit in Geneva to stage a very UN argument about artificial intelligence: can the technology be directed toward public benefit while the infrastructure behind it is increasingly controlled by a small set of companies and rich countries?
WIRED reported that the event filled a 106,000-square-meter convention center near Geneva’s airport district with coding sessions, AI tutorials, panels, Tesla Cybertrucks, UN rescue helicopters, humanoid robots, and other hardware theater. The show floor was loud enough. The policy subtext was louder.
Doreen Bogdan-Martin, secretary-general of the ITU, told attendees that the agency’s belief is that responsibly deployed AI could help address problems including hunger, disease, and climate change. She also said that belief is now being tested by harms created by AI itself.
The summit’s recurring fight was over what “AI for good” means when many of the practical decisions are made inside cloud contracts, chip supply chains, procurement rules, technical standards, and model-release policies. That is less glamorous than a robot dog. It is also where power tends to hide.
Access to compute became the policy fight
Several speakers framed AI access as a development issue rather than a narrow engineering problem. Syed Munir Khasru, chairman of the Institute for Policy, Advocacy, and Governance, said that if AI for good means compute for all, governments and institutions have to treat compute as development infrastructure.
The concern is straightforward enough: advanced AI systems require chips, data centers, cloud services, models, and skilled operators. Countries without that stack can end up renting the future from foreign platforms, with foreign standards attached.
WIRED noted that the debate is taking place while governments are tightening and loosening access to frontier systems. The Trump administration has implemented and then lifted export controls on leading AI models. Reuters has reported that China is considering restrictions on overseas access to its top open-weight models. Poorer countries are the ones most likely to feel the squeeze when access becomes a geopolitical lever.
Speakers also pointed to language as infrastructure. Many large language models are built primarily around English, which leaves smaller languages and local needs underserved. The argument at the summit was that cheaper hardware and smaller local models may matter more for many communities than another frontier demo aimed at wealthy markets.
Rights advocates pushed back on Big Tech dependence
Giulio Coppi, senior humanitarian officer at Access Now, told WIRED that humanitarian and public-sector organizations should stop treating large technology companies as automatic allies. He criticized years of opaque, publicly funded technology deals and warned that some organizations can no longer clearly explain what sits inside their own systems because those systems keep changing.
The event also drew direct protest. Pro-Palestine activists interrupted a keynote by Amazon chief technology officer Werner Vogels, alleging that Amazon technology is being used by Israel against Palestinians, according to WIRED. Security removed the activists from the venue.
Vijay Janapa Reddi, an engineering professor at Harvard University, challenged the vagueness of the summit’s favorite word. He said engineers cannot build toward “good” as a requirement without measurable constraints, comparing it to a plane that only manages to fly for five minutes.
Standards officials made a similar point from a different angle. Gilles Thonet, deputy secretary-general of the International Electrotechnical Commission, told WIRED that engineers have often treated human rights as someone else’s responsibility, a view he rejected. Anja Kaspersen of IEEE argued that consequential AI choices are often embedded in technical architecture, standards, and purchasing decisions rather than debated in assembly halls.
The ITU also promoted a 44-member AI for Good Global Commission, co-chaired by Rwandan president Paul Kagame and Salesforce CEO Marc Benioff. Bogdan-Martin said no single stakeholder can shape AI’s future alone.
That is the consensus version. The harder version is that consensus is slow, procurement is boring, and compute is already being allocated. The robots on the convention floor were a useful reminder that the demos are moving faster than the definitions.
This story draws on original reporting from WIRED.
