multi-agent-systems

Autonomous agents can develop a common communication language through repeated local interactions, without a central computer establishing vocabulary or rules.

A real-world intrusion campaign used hundreds of coordinated AI agents to compress exploit development and mass compromise into machine-speed parallel work.

Social insects provide concrete examples of collectives enforcing system-level constraints through local behaviour even when no individual can measure the state of the whole group.

This paper proposes “Embodied Collective Intelligence,” where robot teams share world memory, task progress, and skill experience without becoming a centralized super-agent.

Pentest Swarm AI uses a shared blackboard, pheromone-weighted findings, and decentralized agent triggers to let attack paths emerge instead of following a fixed pipeline.

This funding call frames AI safety as a population-level problem involving collusion, cascading failures, emergent communication, and collective capabilities. This closes 9-August-2026.

INT21’s PTX Kernel Factory uses specialized agent swarms to generate, test, benchmark, and improve low-level GPU kernels across generations. This is a practical software example of evolutionary adaptation: many agents explore candidate solutions, retain evidence, and feed successful traces into the next cycle.

Google DeepMind’s roadmap treats future AI agents as operational actors that may need containment, monitoring, and layered controls, borrowing ideas from cybersecurity and safety engineering.

DARPA’s new DICE (Decentralized Artificial Intelligence through Controlled Emergence) program may be one of the strongest real-world signals yet that biological coordination principles are moving into operational AI architectures.

AgentSociety proposes a framework where autonomous agents cooperate through mechanisms borrowed from social choice theory and liquid democracy. Agents delegate authority, route information, and form consensus-based decision paths using only local knowledge and incentive-compatible behaviour.

29-Nov-2026 Hong Kong - DAI 2026 places coordination, markets, agent infrastructure, and societies of autonomous agents at the center of distributed AI research.

“Agentic AIs are a feral technology… they are projected to become a wild, self-sustaining, population.” Co-organised with Morgan Stanley, this upcoming Santa Fe Institute gathering takes place 15-Sept-2026 in NYC.