By Marianna Richardson, Director of Communications for the G20 Interfaith Forum and head of its AI working group
———
On 23 June 2026, the Open Source × AI day of UN Open Source Week 2026 — held at United Nations Headquarters in New York and co-hosted by the UN Office for Digital and Emerging Technologies (ODET) and the Office of Information and Communications Technology (OICT) — looked past today’s language models to AI’s next frontier: machines that act in the physical world and software agents that act on our behalf. This summary covers two sessions — a panel on open-source robotics and a discussion on open AI agents — that together made the case for openness as a matter of safety and access, not only economics.
Open-Source Robotics
Physical AI is scaling quickly, the robotics panel observed, and today’s ecosystem is deeply fragmented, with companies producing incompatible hardware and software. As robots move into homes, factories, farms, and public spaces, that fragmentation threatens to deepen inequality — especially if the field remains closed and proprietary. General-purpose robots are already appearing, notably in China, where foundation models power machines that serve coffee and ice cream, work in retail, and perform industrial tasks. In response, some companies have begun open-sourcing the core “brains” of their robots to lower barriers and invite ecosystem-wide innovation, alongside open-source computer vision and specialized systems for agriculture and mining.
Safety dominated the discussion. Unlike a language model, where a wrong answer is usually just inconvenient and a person can catch it, a robot’s mistake can be catastrophic — a difference in stakes that becomes acute as robots enter homes to support aging populations. Panelists argued that safety improves when failures are openly reported, shared, and used to retrain models, drawing an analogy to autonomous driving, where each reported accident can be simulated so that the next generation of software begins from a higher safety baseline. The same logic, they suggested, applies in homes and industry, and points toward robots that are genuinely useful and reliable rather than luxury novelties.
Openness also emerged as the answer to fragmentation. The panel compared robotics to the automotive industry, where enormously varied vehicles nonetheless run on shared infrastructure and common standards. Open source, speakers argued, makes robots easier to update, safer to adapt, and more interoperable across environments; without it, robotics will remain siloed, expensive, and out of reach for most of the world. Simulation-based training and shared learning could extend robotics to developing countries, though the panel acknowledged that the field remains resource-intensive and difficult to deploy, and that true interoperability between different makers’ systems is still largely unsolved.
Open Agents and the Question of Trust
The second session turned to agentic AI — systems that carry out tasks autonomously — and the governance such systems will require. Participants floated the idea of an “AI bill of rights” that would clarify a person’s responsibilities for the agents they deploy, and asked who is accountable when an agent acts. Openness, several argued, is what makes agents auditable: transparency is a problem the open-source world is well suited to solve, with enterprises layering their own security on top of open foundations.
Much of the conversation centered on verification and trust. As the term “open source” is stretched to cover more and more, speakers called for clearer definitions and for verification that can be independently inspected rather than taken on faith — a “proof-of-control” architecture that can confirm an agent actually did what it claimed. Society already has mechanisms for trusting others to act on our behalf, from insurance to legal accountability, and panelists argued that agents will need equivalents in governance, insurance, and accountability. One speaker likened the current moment to “flying the plane while building it”: the technology is moving quickly and remains messy, and it is worth distinguishing what agents should eventually do from what they can reliably do today.
Looking ahead, the panel framed the concentration of AI talent and value in a small number of companies as the central risk of the next five years, and sovereign AI as the opportunity. Rather than routing a population’s intelligence and creativity to a distant hyperscaler, countries could invest in their own systems and adopt a “portfolio” approach that spreads work across many providers and lets local entrepreneurs emerge. Speakers urged that cybersecurity and cryptography be built in from the start, and noted that agentic systems can often run on modest, even out-of-date hardware — an underused resource that could let lower-income countries participate, supported by open compute and university programs. A recurring open question, raised sharply from South Africa, was how to know that data and computation can be trusted when they run on infrastructure owned by others: proving where data resides, how it flows, and that it was computed faithfully remains, the panel agreed, genuinely difficult and unresolved.
———
Marianna Richardson is Director of Communications for the G20 Interfaith Forum, where she leads the Forum’s AI working group and its ongoing exploration of how artificial intelligence intersects with faith, ethics, and public policy. She is also an adjunct professor of management communication at the BYU Marriott School of Business, where she serves as editor-in-chief of the Marriott Student Review and faculty advisor for the Measuring Success Right podcast, and she sits on the International Advisory Council for the International Center for Law and Religious Studies. She attended UN Open Source Week 2026 on behalf of the G20 Interfaith Forum to follow developments in open-source and sovereign AI that bear on the Forum’s work at the intersection of faith and policy.