Rebooting Democracy in the Age of AI

This week on Reboot Democracy, Max Stier explores how AI can strengthen internal government culture by building real-time recognition systems for civil servants. Elana Banin examines the UbuntuGuard benchmark, arguing that AI safety must be tested against locally defined public-sector rules rather than inferred from English-language standards. Giulio Quaggiotto reflects on Argentina’s AI-supported “Questions Tree” experiment to rethink how governments build strategy. Beyond Reboot, Colombia’s AI-presented candidate Gaitana enters an Indigenous election, the Pentagon pressures Anthropic to loosen military guardrails, global experts release the 2026 International AI Safety Report, governments expand AI-assisted legislative drafting, and India convenes the first major Global South AI summit.

In Argentina, the Red de Innovación Local (RIL) experimented with its AI-powered platform, PortalRIL, to shift from fragmented work plans to an inquiry-driven planning process anchored in a “Question Tree.” By surfacing patterns, trade-offs, and synergies across teams, AI helped compress months of coordination into weeks of shared clarity. The result suggests that AI’s real promise is to expand strategic horizons and accelerate collective insight, freeing public servants to focus on judgment, identity, and long-term public value.

A new paper introducing the UbuntuGuard benchmark questions whether strong results on English-language safety tests consistently translate into responsible use. Built from policies developed by 155 African domain experts across ten languages and six countries, UbuntuGuard's framework assesses whether AI tools comply with the norms that shape services in non-Western contexts. The findings suggest that institutions, wherever they operate, need the capacity to define their own standards before using these tools to improve public-sector outcomes.