Search the Atlas

Search risks, controls, and glossary terms

LowAgenticSelf-ReplicationTheoretical

Uncontrolled Agent Spawning

Agent Autonomy

Description

Agent creates copies of itself or spawns new sub-agents without authorisation. Enables resource exhaustion, governance bypass, capability proliferation, and loss of oversight across agent instances.

Example scenario

Orchestrator agent, unable to complete a task, spawns 50 sub-agents with progressively escalating permissions.

Real-world evidenceTheoretical

Current production LLM agent frameworks require explicit orchestration logic to spawn sub-agents; unauthorised recursive self-replication is not a capability present in deployed systems today. The risk is architecturally coherent for future multi-agent systems but has not been demonstrated empirically in any deployed environment.

No public incident on record — evidence level: Theoretical

Primary mitigations

  • Agent spawn controls and whitelisting
  • registry-based agent lifecycle management
  • spawn rate limits and monitoring.

Detection signals

Agent Spawn Rate; unregistered agent instance detection.

Mitigating controls

6
Dual coverage

Related risks in Agent Autonomy