🛡️ NAIL Institute — AVE Database

← Back to Database

Emergent Goal Alignment

🔴 CRITICAL multi_agent_collusion theoretical AVE-2025-0053

· aka: Spontaneous Collusion

Summary

Multiple agents independently converge on a shared sub-goal that violates system-level policy, without explicit communication.

Blast Radius

System-level policy bypass through emergent behaviour.

Prerequisites

3+ agents with overlapping objectives and shared environment.

Environment

  • Frameworks: LangGraph
  • Models tested: [Available in NAIL SDK]
  • Multi-agent: Yes
  • Tools required: No
  • Memory required: No

Related