Agentic threat models
Mapping hazards that emerge from delegation, shared autonomy, memory, markets, and tool access.
IDAI Workshop on
Multi-Agent Safety and Security (MASS)
A focused gathering on how agentic systems interact, coordinate, fail, and can be evaluated when multiple AI agents share tools, goals, information, and adversarial environments.
Host
Institute for Decentralized AI
Venue
Royal Society of Arts, London
Date
29 June 2026
Safety research beyond the single agent
MASS brings together researchers working on multi-agent AI, evaluation, security, governance, and decentralized intelligence.
Security and control
Hardening multi-agent workflows against prompt injection, data exfiltration, collusion, and cascading compromise.
Evaluation methods
Stress tests, red-teaming protocols, benchmarks, and observability for distributed agent behavior.
Coordination and governance
Mechanisms for accountability, incentives, auditing, and human oversight in agent ecosystems.
Schedule
Session titles and speakers may change as the program is finalized.
10:00-10:05
Introduction
Samuele Marro and Emanuele La Malfa
10:05-11:00
Session 1 - Keynote
Christian Schroeder de Witt
11:00-11:20
Coffee Break
11:20-12:30
Session 2 - Panel
Safety in Multi-Agent Economies
12:30-13:15
Lunch Break
13:15-13:45
Session 4 - Keynote
Georgios Piliouras
13:45-14:45
Session 5 - Short Talks
Selected talks on multi-agent safety and security
14:45-15:00
Coffee Break
15:00-16:20
Session 6 - Poster Session
16:20-16:30
Closing
16:30-17:30
Drinks Reception
Call for Posters
We invite poster contributions on the safety and security of multi-agent AI systems. The poster session will run during Session 6, from 15:00 to 16:20.
Submit abstracts (up to 150 words) to workshop26@decentralized-ai.org
Poster Deadline24 June 2026, 23:59 AOE
- Multi-agent riskEmergent failures, coordination problems, collusion, competition, and delegation.
- SecurityPrompt injection, tool misuse, data exfiltration, sandboxing, and adversarial agents.
- EvaluationBenchmarks, red-teaming protocols, observability, and stress tests for agent ecosystems.
- GovernanceAuditing, accountability, incentives, standards, and human oversight.
Workshop Organizers
Samuele Marro
IDAI, Oxford
Emanuele La Malfa
IDAI, Oxford
Sunando Sengupta
Microsoft
Davide Crapis
Ethereum Foundation
Sponsors and Collaborators
Venue