What Agent Emergency Stop (Kill-Switch) Design Patterns requires
An AI Agent Kill-Switch is a deterministic safety mechanism designed to immediately terminate or throttle an autonomous agent's execution if it exceeds predefined behavioral, financial, or operational boundaries. A compliant kill-switch architecture requires: sub-50ms signal propagation to prevent runaway execution, a graceful shutdown window with hard-terminate fallback on timeout, state snapshot capture before forced termination, dead letter queue routing for incomplete tasks, rollback of reversible actions, mandatory human-in-loop approval before restart, and audit logging of every kill event with trigger classification. The corrigibility principle underlying kill-switch design - that agents must remain stoppable and correctable by authorized humans - is foundational to EU AI Act Article 9 risk management requirements and to NIST AI RMF MANAGE 4.1 incident response protocols.
Pillar: Workflow Automation · Authority: National Institute of Standards and Technology (NIST) · Version: 1.1.0 · Last updated:
Primary source: https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.100-1.pdf
SHA-256 integrity: 3f810fa71ac533c7491ee80e1853007adbda8bd555233b0dc11e33619ac64ea2
Primary Citations — 7 traced to source
- Stuart Russell - Human Compatible AI (2019) - Chapter 5 (Corrigibility and Safe Interruptibility)
- Hadfield-Menell et al. - The Off-Switch Game (IJCAI 2017)
+ 5 more citations (full bibliography, deterministic workflow, actionable schema and crosswalks) included in the vault unlock — $0.01 via Skyfire / L402 / Direct Base USDC.
Access
- Discovery (free): /api/v1/nodes/agent-kill-switch.json — 6-field metadata
- Vault (full node): /api/v1/vault/nodes/agent-kill-switch.json — full 13-key payload, $0.01 USDC (L402/Skyfire/Direct Base)
- Canonical URL: https://bidda.com/intelligence/agent-kill-switch
- Back to registry: Browse all 10,085 compliance nodes