Responsible AI Takes Shape

This week, we convened our global community of partners in Geneva, alongside the first UN Global Dialogue dedicated to AI governance and the release of the UN international scientific panel’s evidence-based assessment for world leaders. As PAI enters our next decade, the 2026 Partner Forum reflects a shift we’ve been building towards—taking responsible AI from principle to practice, at scale, in close partnership with the organizations doing this work every day. We were excited to announce two new initiatives at the Forum: the Global AI Progress Hub, a public tracker where organizations across sectors can share the actions they’re taking, the progress they’re making, and the outcomes they’re producing, and the Global Responsible AI: Measures of Progress Report, which will complement the hub as an independent, annual assessment of whether those actions are producing real-world results. Across the evening’s panel discussions, a fireside chat and a keynote, three throughlines kept surfacing: the impacts of AI today, who is accountable for it, and who gets a seat at the table as those questions get answered. “AI is generating genuine value, though where that value lands still depends on who’s using the tool and with what support.” What AI Is Delivering Today Speakers at the Forum offered a look at AI’s impact right now. One organization described handing off conversations from an AI system to a trained human volunteer, overseen by mental health professionals, supporting more than 17 million conversations, with many users under 18. It’s a strong illustration of AI’s potential and a reminder of how much capacity building still has to happen alongside it, a tension we’re increasingly focused on in PAI’s own work on AI and human connection. A different kind of impact we’re seeing is leading research labs’ contributions to drug discovery as a case where AI is producing real, measurable scientific value rather than speculative promise. With labor and the economy, the impact is mixed. Call center productivity has climbed 14% -15% with AI assistance, yet wages haven’t moved in step. Algorithmic management has also added new pressure to jobs rather than easing it. Meanwhile, a PwC study estimated assistive AI generated up to $330 billion in socioeconomic value for households through tools that help manage childcare, eldercare, and daily logistics, along with data showing productivity gains for small businesses. Together, these findings suggest AI is generating genuine value, though where that value lands still depends on who’s using the tool and with what support. The Case for Existing Frameworks A recurring argument across the Forum, one we’ve heard echoed across our broader partner community over the past year, is that before reaching for new governance mechanisms, it’s worth looking harder at what’s already in place. The UN Guiding Principles on Business and Human Rights framework predates generative AI but still holds up well today, laying out clear obligations for governments, clear responsibilities for companies, and a strong foundation for multistakeholder engagement. The real gap here is remediation: many companies haven’t formally committed to the UNGPs, and enforcement remains inconsistent even among those that have. Another speaker noted that US consumer protection regulation may already cover agentic AI payments conceptually, and that building the technical infrastructure to make that coverage work in practice matters more than drafting new law. Partnership Means More Seats at the Table Another resonant theme was who gets included in shaping AI, and where that inclusion is still falling short. Workers rarely have real influence over how AI gets deployed in their own workplaces. But there are strong examples of workers successfully influencing how AI is used. A writers’ guild negotiated AI protections through collective bargaining, and a banking sector oversight committee in Europe now reviews AI-driven job cuts before they take effect. Both are the kind of outcomes we’ve long tried to support through our own guidance for AI and shared prosperity. The same question showed up again at the country level. PAI’s scenario-planning work is finding that while nearly everyone agrees AI will reshape economies, optimism tracks closely with a country’s growth trajectory. Where growth has stalled, AI often reads as an opportunity to reset it, while in more established labor markets, it reads mostly as risk. Closing that gap requires making sure AI’s benefits accrue directly to the countries and communities most affected, not only to the ones already ahead—a distinction central to PAI’s planned international expansion in the coming years. “Reaching more voices. . . means thinking carefully about where and how we bring people together.” Access is critical to how we think about convening our own community going forward. Global AI governance gatherings, including our own, often strive to be inclusive in principle but remain prohibitive in practice for civil society and Global South participants, clustering in expensive, hard to reach cities. Reaching more voices, including communities working on low-resource languages who are too often left behind, means thinking carefully about where and how we as a community of thought leaders bring people together. Building What Comes Next In our closing conversation, we heard about Switzerland’s own model of governance: participatory, multistakeholder, and deliberately unhurried. This offers a working example of the kind of inclusion the rest of the evening had been asking for. Solutions built this way tend to be ones people actually want to implement, rather than ones they’re pushed into following. We’re excited to see how this shapes the 2027 AI Summit in Geneva. With a partner community now spanning more than 150 organizations across 20-plus countries, our focus going forward is to make sure that what our community is doing, and the difference it’s making, is visible, evidenced, and built on year over year. To stay up to date on our work, sign up for our newsletter.