When AI Agents Meet: A Blueprint for Addressing Multi-Agent AI Governance
In partnership with the Cooperative AI Foundation (CAIF), "When AI Agents Meet: A Blueprint for Multi-Agent Governance" was held virtually on 10 September 2026. Lewis Hammond (Research Director, CAIF) and Elham Tabassi (Director, AI and Emerging Technology Initiative; Senior Fellow, Brookings) open by mapping where we are: the early stages of watching AI agents come into contact with one another, with governance instruments that currently assume the unit of assessment is a single system and the unit of accountability is a single organization. A panel moderated by Marta Bienkiewicz (Policy and Partnerships Manager, CAIF) then works through the five action points proposed in the blueprint "Establishing Foundational Principles and Thresholds for Multi-Agent AI Governance," which came out of the CAIF workshop at IASEAI’26. Evaluate — Patricia Paskov (Director of Standards, AVERI) and Sebastian Hallensleben (Chair, CEN-CENELEC JTC 21; OECD) on multi-agent failure modes, independent third-party evaluation, and what AI can learn from how aviation treats incidents Identify — Sam Boger (Research Fellow, AI Agents, Singapore AI Safety Hub) on agent IDs as flexible containers for attribution, accountability, and incident response Monitor — Nada Madkour (Director, AI Security Initiative, CLTC Berkeley) and Patricia Paskov on interoperable logging and keeping human oversight meaningful at agentic speed Report — on near misses, cross-provider formats, and the limits of voluntary disclosure Incentivize — Sebastian Hallensleben on putting accountability in the same hands as control across a long value chain The session closes with an audience Q&A that touches on cross-border incidents, agents communicating in ways humans can't read, and insurance as a mechanism for pricing AI risk. The speakers are clear that this is a starting point rather than a finished framework, and that the work is both tractable and urgent. Learn more and get involved at iaseai.org!