lightningFraming & overviewCross-cutting

Safety for the First Billion Agents

8 July 2026 · 11:49 am–11:54 am · Cullen

Current AI safety is not enough to prevent extreme power concentration, gradual disempowerment, and resource contention with social-scale agentic systems. Safety needs a new approach which acknowledges the unique challenges and tooling of large agent systems, from efficiently scaling existing techniques to developing infrastructure which adapts to its participants. I will discuss new demands of large agent systems, compare nascent approaches, and note barriers to progress. I will conclude by sharing a community and centralised resources for interested researchers.

Speaker

S
Stephen Elliott
Research Director, Gigascale Laboratories
Join the large agent systems slack community: https://join.slack.com/t/largeagentsystems/shared_invite/zt-414xse5s4-K5GY6kNFN7IkPV5OdIDvTg I started Gigascale Labs in February to address the gap in large agent systems research. We model AI influence in social network science, financial systems, macroeconomics, and politics, to understand gradual disempowerment as it occurs in the real world. We're building efficient open-source research tooling to extract agent profiles from real large systems datasets, grounding simulations in real-world behaviour. Please reach out!

Audience Q&A

Ask a question or upvote others.

Loading questions…

View full program