AI safety: From the Fence to the Field

Despite the popular notion, AI safety is close to the opposite of doom-luddism; many of the same ideas and people behind it also built the AI breakthroughs we have today.Ankur will cover various tribes & camps in AI safety and how the same safety/alignment problems are interpreted differently by various camps. He’ll get into underserved areas and blind spots where India can play a role - a strong build-stuff culture, representing ‘middle powers’, etc. Ankur will cover what one can do in AI safety today - technical work is only one type, and despite the name, governance work can be pretty technical. He’ll cover the intellectual predecessors of AI safety - Effective Altruism, Rationality, etc. He’ll also discuss economics and the job market in AI safety - how (outside big labs) it’s largely philanthropy driven. And that (a lot of) AI safety should be treated as a product with real stakeholders, incentive structures, etc. About the speaker: Ankur Pandey was first introduced to proto AI safety (then called Friendly AI) through interactions with MIRI, the Centre for Effective Altruism, and Singularity University - about 15 years back. At the time, it felt like worrying about overpopulation on Mars.He spent the years in between in applied ML and then as a founder across three bootstrapped startups, including LongShot AI. In 2025, he realized that watching AI from the fence was no longer exciting and made a hard pivot into AI safety and adjacent work. Since then, he has been an active field builder (organized an international book club on AI + society, events at the India AI Impact Summit, etc). Did multiple cohorts (BlueDot, Center for AI Safety, Cambridge Digital Minds). Been a fellow at SPAR, AI Safety Camp, and an invited guest at the PIBBSS workshop & EA global. He also consulted for Alice (AI security). He currently consults / contributes for Groundless.ai on AI epistemic risk and continues independent AI safety research. Most importantly, he is building a fellowship around AI safety for builders (backed by grants from Manifund). Pre-read: Priority AI risks - MIT survey of 272 AI experts ranking which risks are most severe and who's responsible for them SaferAI Frontier Risk Management Tracker The Case for AGI Safety Products by Apollo Research Join online: meet.google.com/zuc-kkdf-bxc Looking forward to seeing you!