Agentic artificial intelligence, better known as AI Swarms, has advanced to an incredible level of independence, reasoning, and coordination. This raises fears about their malicious capabilities, e.g., influencing public votes, manipulating online discussions, organizing political campaigns, or inciting violence.

To better understand and predict their collective behaviour needs a joint scientific effort. We can build on recent advances in different disciplines, but need to unite their contributions for a comprehensive understanding of AI swarms.

The collective behavior of AI swarms can both lead to malicious uses and to unintended consequences in benign applications where systemic risk can emerge. The availability of AI agents for experimentation and benchmarking is a unique opportunity to inform Complexity Science approaches that can explain swarm behavior and motivate interventions and policies.

• Understanding: To what extent can LLM-based multi-agent systems exhibit swarm behavior?
• Interventions: How can malicious AI swarms be detected, steered, or disrupted?
• Methodologies: How can we empirically understand the individual and collective behavior of AI agents?

Please note that participation is by invitation only.