Jekyll2026-09-01T06:01:12+00:00https://www.gigascale-labs.org/feed.xmlGigascale LaboratoriesGigascale Laboratories' websiteGigascale LaboratoriesLargeAgentSystems.org, a Resource for Large Agent Systems Safety2026-09-01T00:00:00+00:002026-09-01T00:00:00+00:00https://www.gigascale-labs.org/posts/2026/09/largeagentsystems-orgWe are pleased to share largeagentsystems.org, a new resource we have been working on for the last couple of months. It presents resources for safety on large groups of agents.

The site has a problem definition, an org map, a survey, a daily paper scraper, and a weekly events scraper. Together these help develop a common understanding and a basic resource pool for large agent systems safety problems.

Large agent systems, or systems with thousands to billions of agents, are important because:

  • They have shown us the first takeover-type event, in the OpenAI Hack.1 Apparently the emergent capabilities and agency discussed in Multi-Agent Risks2 are more immediate than other safety areas realised.
  • They overlap heavily with human systems. The first really big systems of agents are economic and social systems, where AI mixes in with and slowly replaces humans. This makes large agent systems safety the principal setting for disempowerment, job loss, inequality, and similar harms.

Large agent systems have many different properties to smaller-scale multi-agent systems, particularly observability, decentralisation, scalable monitoring, complex coevolution of agent behaviour and system mechanisms, and a lack of datasets to validate our understanding against.

Further information is available on the site, at largeagentsystems.org.

References

  1. METR. 2026. Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident. METR blog, 26 August 2026. https://metr.org/blog/2026-08-26-openai-hugging-face-incident-investigation/ ↩

  2. Lewis Hammond, Alan Chan, Jesse Clifton, Jason Hoelscher-Obermaier, Akbir Khan, Euan McLean, Chandler Smith, and others. 2025. Multi-Agent Risks from Advanced AI. arXiv preprint arXiv:2502.14143. https://arxiv.org/abs/2502.14143 ↩

]]>Stephen Elliotthttps://stelliott.online