AI for Good Global Summit 2026

About

The AI for Good Global Summit is ITU's flagship gathering on applying artificial intelligence to the Sustainable Development Goals. It draws thousands of participants (governments, UN agencies, researchers, companies, and civil society) from around 170 countries.

The 2026 edition anchored a wider Geneva “AI week” in July, running alongside the WSIS Forum and immediately after the inaugural UN Global Dialogue on AI Governance.

What happened

ITU announced the AI for Good Global Commission five days before the summit opened: a body of more than 40 founding members co-chaired by Rwandan President Paul Kagame and Salesforce chair Marc Benioff, with ITU Secretary-General Doreen Bogdan-Martin as vice-chair. It seats heads of state (Estonia, Iceland, Kazakhstan) alongside the executives of the firms building the technology (Amazon, Anthropic, Cohere, Microsoft, Nvidia), with a mandate to widen access, strengthen trust and accelerate impact. African representation runs through it, with the African Union seated alongside Namibia, Nigeria and Togo.

Switzerland also used the week to announce that Geneva will host a recurring World Summit on AI from 2027, positioning the city as the standing venue where the AI-and-development agenda is negotiated.

The summit drew its largest audience yet, with organisers and press reporting more than 12,000 participants from 170 countries. Much of the week ran through its competition tracks: Nearpays won the Innovation Factory, with will.i.am announcing additional funding for all three finalists; Plastic Erosion took the AI for Good Film Festival; the Robotics for Good Youth Challenge brought roughly 250 children to the venue; and the third day opened with International Girls in ICT Day. The exhibition floor carried the summit’s assistive technology case, from brain-computer interfaces to a $4,000 robotic guide dog shown by China Mobile.

The most consequential thing the summit produced was also the least photogenic. On 9 July the ITU announced a Focus Group on Trust and Identity for Humans and Agentic AI, a standards-track body addressing what happens when AI stops advising and starts acting: agents that impersonate a person or an organisation, or that take unauthorised actions across connected systems. It will write the terminology, reference architectures, trust frameworks and interoperability mechanisms, with the explicit aim of preserving meaningful human control over things like financial transactions and critical infrastructure. Co-chaired by Debora Comparin and Amir Banifatemi, it reports to ITU-T Study Group 17 and meets first in Paris in November 2026, then Geneva in January 2027.

This is where the summit’s week connects to the rest of Geneva’s. The Global Dialogue on AI Governance set the political frame for who governs AI. A focus group on agent identity is the unglamorous machinery that decides whether any of it is enforceable, because a rule that cannot establish which agent acted, on whose authority, is not a rule.

The summit closed on 10 July, with ITU framing the final day around women’s leadership, education and human progress.

This section draws on press coverage of the summit rather than our own session transcripts, because the summit does not stream publicly and publishes no official transcripts. Attendance figures are as reported by organisers and press. Primary sources are linked under Key links below.

Who gets paid for the training data

The sharpest economic argument of the week came from the music track, and it came in a keynote of its own. On 8 July ITU gave Björn Ulvaeus, the ABBA co-founder who now presides over CISAC, the global confederation of authors’ societies, ten minutes under the title “The future needs creators”. A separate panel the next day, “The Sound of Intelligence,” put John Legend alongside executives from Universal Music, Stability AI and Nvidia. Ulvaeus’s move was to reject the terms on which the copyright fight has been conducted.

The industry has spent three years arguing about outputs, trying to trace whether a generated song copies a particular recording. Ulvaeus argued that this was always the wrong question, because a model does not copy songs, it produces a new synthesis of everything it absorbed. If the output cannot be traced back, the claim has to be made at the other end, on the input.

The summit streams nothing and publishes no transcripts, and there is no recording of this keynote to draw on. The quotation is as reported by the music trade press, which covered the keynote closely and has its own interest in the claim being made.

What we were watching, and how it landed

• Applied AI at scale. The summit did lean into deployments over abstract debate, with 300+ sessions and an exhibition floor running from brain-computer interfaces to a robotic guide dog. The durable output, though, was a standards body on agent identity, not a demonstration.
• From hype to governance. Following straight on from the Global Dialogue, the “AI for the SDGs” optimism did sit alongside harder questions about oversight and access. The Global Commission is where that tension now lives: it seats heads of state next to the companies building the technology, and whether it can hold both to account is unresolved.
• Who gets to watch. Paid registration and no public livestream held for 2026, at an event built around inclusion. The record is the recordings posted afterwards and ITU’s own daily recaps, which is a thinner public account than the co-located UN tracks left behind.

Session coverage

The summit does not stream publicly and publishes no official transcripts. These summaries are built from the recorded sessions posted afterwards, so they arrive later than our coverage of the webcast UN events, and quotations are close paraphrase from machine transcription.

Why it matters for the SDGs

The week’s argument about training data is where the summit’s premise meets SDG 3 and SDG 10. A model that answers a maternal-health question worse in Kiswahili than in English underserves the language’s speakers, and no connectivity statistic registers it: the person counts as online, and is served badly anyway. That is a gap the SDG indicators are not currently built to see.

The institutional answers the week produced pull toward SDG 9 and SDG 17. The ITU’s Focus Group on Trust and Identity for Humans and Agentic AI is the machinery question: as AI moves from advising to acting, a rule that cannot establish which agent acted, on whose authority, is not enforceable, and that reaches critical infrastructure and financial systems directly. The AI for Good Global Commission is the partnership question, and an unsettled one. It widens the table to the African Union, Namibia, Nigeria and Togo alongside the largest AI firms, which is either the point of SDG 17 or the limit of it, depending on whether a body that mixes governments and industry can hold industry to account.

Important documents

The summit's flagship reports on AI and the Sustainable Development Goals.

Watch & read

The summit published no outcome document or proceedings. The public record is the recordings posted after the event on the AI for Good YouTube channel and the summit’s “Neural Network” platform, plus ITU’s own announcements and daily recaps.

• AI for Good on YouTube, session recordings
• AI for Good Summit 2026, official site
• AI for Good Global Commission, ITU launch announcement (2 July 2026)
• Pope Leo XIV’s message to the summit (8 July 2026)
• ITU Focus Group on Trust and Identity for Humans and Agentic AI, launch announcement (9 July 2026)
• AI for Good Global Commission, the full list of commissioners
• AI for Good newsroom, ITU’s daily recaps and announcements
• ITU’s closing recap, the official summary of the summit’s final day
• “The Sound of Intelligence”, the session on AI, music and creators’ rights
• WSIS Forum 2026 · UN Global Dialogue on AI Governance, the co-located tracks