ELIAS @ NeurIPS2026 | Foundations of Agentic Systems Theory (FAST) workshop

Exploring the foundations of agentic AI, FAST brings together interdisciplinary perspectives to understand emergent behaviors, collective intelligence, and the capabilities and risks of interacting AI agents.
12 December, 2026

The Foundations of Agentic Systems Theory (FAST) workshop will take place on December 12, 2026 in Paris, held at NeurIPS 2026. FAST brings together researchers from AI, complex systems, developmental biology, organizational sociology and cognitive science to ask how the behavior of systems of LLM-based agents emerges, and how it can be understood and controlled.

12 DecWorkshop day, 2026
ParisPalais des Congrès
4Keynote speakers
8Research topics

About the workshop

The most interesting and consequential behaviors of complex systems often arise not from the parts in isolation, but from the patterns of interaction between them. Current development of agentic AI has largely focused on building more capable individual agents, which risks underestimating both the capabilities and the risks of agents once they become widespread.

Existing theory from control, game and economic theory, operations research and the social sciences does not map cleanly onto agentic AI: LLM-based agents are neither idealized rational agents nor culturally shaped human agents. FAST provides a venue to investigate which mechanisms of emergent behavior carry over to systems of LLM-based agents, which properties of the underlying agents facilitate or impede them, and how far system-wide outcomes can be controlled or induced. Interdisciplinary participation is strongly encouraged, through both contributions and invited talks.

Topics of interest

Mechanisms of emergent capabilities and behaviors in agentic systems
Evaluation, detection, and bounding of emergent capabilities or failure modes
Harness engineering and (neuro-symbolic) scaffolding
Theory of mind and recursive social cognition
Formation of norms, conventions, and collective bias in populations of agents
Compositional safety and governance
Definitions and philosophy of agency and emergence in engineered systems
Observability, monitorability and controllability of populations of agents

Keynote speakers

Michael LevinTufts UniversityStudies bioelectricity, morphogenesis and collective cognition, and how competent parts compose into collective agents with goals of their own.
Danielle PerszykAmazon AGI SF LabCognitive scientist leading the human-computer interaction team, working on practical AI agents that act in digital and physical environments.
Winnie StreetGoogle ResearchResearches social cognition in large language models, including theory of mind and its implications for alignment and group dynamics.
Atoosa KasirzadehCarnegie Mellon University / Google DeepMindPhilosopher of AI ethics and governance, working on AI existential risk and on characterizing AI agents for alignment and governance.

Program

The workshop opens with brief remarks from the organizers, followed by a mix of keynotes, contributed talks, poster sessions and a panel, with networking breaks throughout to maximize discussion time.

Keynotes30-minute keynote talks from leading researchers across cognitive science, biology, philosophy and AI.
Contributed talks and postersTwo contributed talk sessions and two poster sessions presenting accepted papers.
Panel discussionAn hour-long panel, “Mind to machines: Emergence, methods, and risks”, moderated by Joshua Krook. Panelists will be announced closer to the workshop.

The detailed schedule will be posted on the workshop website once paper decisions have been finalized.

Important dates

September 5, 2026 (AoE)Paper submission deadline
September 29, 2026 (AoE)Acceptance notification
December 12, 2026Workshop
December 20, 2026 (AoE)Camera-ready deadline

Submission information

Full papersUp to 7 pages, excluding references and appendices. For mature or completed research.
Short papersUp to 4 pages, excluding references and appendices. For ongoing work, early-stage ideas, or the release of benchmarks and datasets.

Submissions are made through OpenReview using the NeurIPS 2026 template, and must be anonymized for double-blind review. Accepted papers are presented as posters or contributed talks. Papers are non-archival, so work under submission elsewhere is welcome.

Registration

Registrations for the NeurIPS Europe event have sold out, and a waitlist is open. Registration is complimentary for financial aid recipients, volunteers, top reviewers, ACs/SACs, and authors of papers at affinity events. A second call for volunteers, open to PhD students and postdocs and including free registration, will open in early October. The venue is the Palais des Congrès de Paris.

Organizers

Erik MiehlingIBM Research - Ireland
Madeline ReineckeUniversity of Oxford
Hamza MostafaUniversity of Waterloo
Jordan McAfooseIBM Research - Zurich
Irina RishUniversité de Montréal / Mila
Djallel BouneffoufIBM Research - Yorktown Heights
December 12, 2026 · Paris. Foundations of Agentic Systems Theory at NeurIPS 2026. Questions? Write to fast.workshop.team@gmail.com or visit the workshop website.

Upcoming Events