AIOGF‑SD‑12.2.2 — Kill Switches

Document Identifier: AIOGF‑SD‑12.2.2
Related Control: 12.2.2
Framework: AI Operational Governance Framework (AIOGF)
Author: Randy Manthey
Version: 1.6
Date: March 24, 2026
Status: Working Draft
© 2025–2026 Randy Manthey. All Rights Reserved.

Licensing and Usage Notice

This supplemental document is part of the AI Operational Governance Framework (AI-OGF) and is protected under the AI‑OGF Limited Use License.

You may:

  • Read and reference this document for internal, non‑commercial use.

You may not:

  • Reproduce, redistribute, or create derivative works.
  • Use this document for commercial purposes, consulting, training, or resale.
  • Use this document to train AI models or automated systems.
  • Incorporate this document into tools, platforms, or governance products without written permission.

For permission requests or collaboration inquiries, visit the Permission and Collaboration page on the official AI-OGF site.

12.2.2.1 Purpose of the Practice

The purpose of this practice is to ensure that AI systems can be immediately and irreversibly terminated through a kill switch mechanism when behavior becomes unsafe, uncontrollable, or catastrophic.

Kill switches provide the highest level of human authority and are reserved for emergency conditions.

12.2.2.2 Scope & Applicability

This practice applies to:

  • AI systems with high‑impact or destructive capabilities
  • AI systems with privileged access
  • AI systems controlling infrastructure, identity, or security
  • multi‑AI environments where cascading failures are possible

Organizations must implement kill switches that immediately terminate AI execution and prevent further action under emergency conditions.

12.2.2.4 Rationale

Kill switches are required because AI systems may:

  • enter runaway workflows
  • exceed autonomy boundaries
  • cause cascading failures
  • become unresponsive to override
  • act in ways that threaten safety or continuity

Kill switches provide a final, non‑negotiable safety mechanism.

Foundational Principle

AI systems must be terminable at any time through a human‑controlled kill switch, regardless of operational state or autonomy level.

12.2.2.5 Implementation Guidance

Organizations should:

  1. Implement kill switches at the system and workflow level
  2. Ensure kill switches terminate execution immediately
  3. Prevent AI systems from disabling or bypassing kill switches
  4. Require human approval to restore service after activation
  5. Log all kill switch activations

12.2.2.5.1 Preconditions

  • identity and access controls
  • kill switch authority definitions
  • workflow instrumentation

12.2.2.5.2 Scope & Impact Analysis

Evaluate:

  • which systems require kill switches
  • which actions pose catastrophic risk
  • which dependencies must be terminated

12.2.2.5.3 Standards Alignment

Aligns with:

  • safety‑critical system design
  • operational control
  • risk management

12.2.2.5.4 Trust Relationship Evaluation

Evaluate:

  • cross‑AI workflows
  • privilege propagation
  • identity dependencies

12.2.2.5.5 Privilege Escalation Assessment

Assess:

  • whether AI can disable kill switches
  • whether kill switches can be misused

12.2.2.5.6 Automated Validation

Automated systems should:

  • detect kill switch activation
  • pause dependent workflows
  • alert operators

12.2.2.5.7 Human Review Requirements

Human review is required for:

  • kill switch activation
  • system restoration
  • post‑incident analysis

12.2.2.5.8 Downstream Impact Analysis

Evaluate:

  • impact on dependent systems
  • impact on identity and access
  • impact on cross‑AI interactions

12.2.2.5.9 Documentation Requirements

Document:

  • kill switch events
  • root cause analysis
  • corrective actions

12.2.2.6 Business Impact

Failure to implement kill switches may result in:

  • uncontrolled AI actions
  • catastrophic operational failures
  • irreversible changes
  • governance violations

12.2.2.7 Expected Outcomes

Organizations should expect:

  • immediate termination of unsafe AI behavior
  • improved safety and control
  • reduced risk of catastrophic failures

12.2.2.8 Examples

Example 1 — Runaway Workflow

An AI system enters a recursive loop; the kill switch terminates execution.

Example 2 — Destructive Action Attempt

An AI system attempts a high‑risk destructive action; the kill switch halts the system.

12.2.2.9 Alignment to External Frameworks

NIST AI RMF — Govern (extension)
ISO/IEC 42001 — Operational Control (extension)

12.2.2.10 Notes

Kill switches must remain functional even in degraded or failure conditions.

12.2.2.11 Cross‑References

Internal AI-OGF Controls:

  • 12.2.1 Manual Override
  • 12.2.3 Human‑Validated Checkpoints
  • 8.4.x Override & Rollback

Additional technical guidance, architectural patterns, and diagrams for kill switch implementation are provided in:

  • Annex K — Kill Switch Architecture & Safety Interlocks (eta-tbd)
  • Diagram Set K — Kill Switch & Safety Interlock Visuals (eta-tbd)

This document is part of the AI Operational Governance Framework (AI-OGF) and is protected under the AI-OGF Limited Use License. Official source: https://rmanthey-mantheyco.github.io/ai-ogf/

This site uses Just the Docs, a documentation theme for Jekyll.