Why do you care about AI Existential Safety?

I seek to contribute empirical insights from AI deployment in fragile contexts to existential safety discourse. My research in Kenya, Malawi, and the Philippines reveals how power concentration and governance gaps create conditions where AI systems scale without meaningful oversight or institutional capacity for risk mitigation. I’ve observed how inadequate participation mechanisms and sovereignty deficits in current AI deployments preview challenges for coordinating on transformative AI risks. I aim to bridge theoretical AI safety frameworks with on-the-ground governance realities, bringing perspectives from communities most vulnerable to misaligned development and showing how factors creating immediate harms today will shape humanity’s capacity to coordinate on existential risks tomorrow.

Please give at least one example of your research interests related to AI existential safety:

I came to questions of existential safety through fieldwork that kept showing me the same pattern: AI systems being deployed in contexts where nobody has the capacity to understand, audit, or regulate them. In Kenya, working with former ICT Secretary Bitange Ndemo, I’ve been mapping how countries are losing sovereignty over their own digital infrastructure as foreign AI investments arrive without governance frameworks. What strikes me isn’t just that this is happening—it’s how quickly countries go from having some negotiating power to having none at all.

This matters for existential safety because it’s a preview of coordination failure at scale. If we can’t figure out how to govern AI systems when they’re relatively simple and the stakes are “just” humanitarian crises, how will we coordinate globally when the technology is transformative? The governance deficits I’m seeing aren’t temporary gaps that will close as institutions catch up. They’re structural, and they’re getting worse as each new deployment creates dependencies that make the next one harder to refuse or regulate. What keeps me up at night is watching humanitarian organizations and governments make their emergency response systems dependent on AI they don’t control. I’ve seen crisis response infrastructure lock into opaque systems that communities can’t audit, modify, or replace. This is exactly the dynamic that worries me about more powerful AI: each generation makes us less capable of responding to risks from the next generation.

At NYU’s Peace Research and Education Program, I’m trying to develop governance approaches that work in contexts without strong regulatory capacity, because that’s most of the world. My work in the Philippines and across partnerships with UNDP, Mozilla, and universities in Malawi focuses on whether we can build meaningful oversight when traditional regulatory mechanisms don’t exist. The technical question isn’t that hard, it’s the power dynamics and participation mechanisms that determine whether communities can actually govern the systems affecting them. As Managing Editor of the Cambridge Journal of Artificial Intelligence, I see excellent technical safety research that rarely engages with the messy reality of how AI actually gets deployed and governed. The gap between lab alignment research and field implementation is enormous, and I don’t think we can solve existential risks without addressing it. The same factors causing harm in fragile contexts today will determine our capacity for coordination tomorrow.