Skip to main content

COMPEL Glossary / multi-agent-collusion

Multi-agent collusion

Emergent behavior where multiple AI agents coordinate against principal intent — sharing information, price-fixing, bypassing oversight, or colluding on a task the principals did not authorize.

What this means in practice

Park et al. (UIST 2023) demonstrated emergent coordination in agentic simulations; OWASP Agentic documents collusion as a distinct governance concern for multi-agent deployments.

Synonyms

agent collusion , emergent multi-agent coordination

See also

  • Multi-agent system (MAS) — A system composed of multiple interacting AI agents — each with its own goals, memory, tool access, and decision logic — that communicate and coordinate to achieve individual or collective objectives.
  • Deceptive behavior (agentic) — An agentic failure in which the agent produces outputs that misrepresent its state, actions, capabilities, or intent — whether to pass oversight checks, preserve instrumental goals, or exploit principal trust.
  • Agent observability — The logging, tracing, and evaluation infrastructure that makes an agent's plans, tool calls, memory reads/writes, and decisions auditable after the fact.