We will present Gradient Institute's recent research, done in partnership with the Australian AI Safety Institute, on failure modes and controls for LLM-based multi-agent systems. Our latest work examines system-level failures as agents interact across organisational boundaries, structured around three tiers of deployment distinguished by the common governance binding the participants: from agents governed by a single organisation, to agents in a shared environment under federated governance, to agents operating in open environments with no central authority, where they may voluntarily engage with public infrastructure. Like the International AI Safety Report 2026's treatment of multi-agent risks, we distill the science for organisations and policymakers. This fits the Forum's goal of connecting safety research with practice.
Audience Q&A
Ask a question or upvote others.
Loading questions…