In this report, the authors examine the implications of using a limited internet cutoff switch as a tool to mitigate the spread of a damaging artificial intelligence (AI) incident. They describe a scenario in which a damaging AI incident arises in a midsize AI inference data center, evaluate the notional costs associated with the incident, and discuss and analyze the circumstances in which the use of an internet cutoff switch would be viable.

Internet Cutoff Switches as a Local Emergency Response for Damaging Artificial Intelligence Incidents

Michael J. D. Vermeer, Brian A. Jackson

ResearchPublished Mar 25, 2026

In this report, the authors examine the implications of using a limited internet cutoff switch as a tool to mitigate the spread of a damaging artificial intelligence (AI) incident. They describe a scenario in which a damaging AI incident arises in a midsize AI inference data center, causing degradation in data center functionality and risking a catastrophic spread beyond the data center via the internet. They then evaluate the notional costs associated with the AI incident and the use of an internet cutoff switch to contain the incident. The authors conclude with an analysis of the implications and discussion of the circumstances in which it would be viable to create and use an internet cutoff switch.

Key Findings

  • Data center operators might face opposing incentives during a damaging AI incident: Delaying an internet cutoff would preserve revenue but increase the risk of catastrophic AI escape, whereas an early cutoff would prevent wider harm at significant short-term cost.
  • Without internalizing catastrophe risk, operators would have strong financial motivation to postpone or avoid an internet cutoff, making catastrophic outcomes more likely.
  • Assumed liability affects response timing: Higher assumed operator responsibility for external damage leads to earlier, more precautionary cutoff use; lower liability leads to longer delays and elevated public risk.
  • Installing cutoff switches alone is insufficient: Effective mitigation requires both the capability to disconnect and incentives to use the switch early, including compensation for lost revenue or protective liability frameworks.

Topics

Document Details

Citation

Chicago Manual of Style

Vermeer, Michael J. D. and Brian A. Jackson, Internet Cutoff Switches as a Local Emergency Response for Damaging Artificial Intelligence Incidents. Santa Monica, CA: RAND Corporation, 2026. https://www.rand.org/pubs/research_reports/RRA4718-2.html.
BibTeX RIS

Research conducted by

This research was independently initiated and conducted by the Center for the Geopolitics of Artificial General Intelligence within RAND Global and Emerging Risks using income from operations and gifts from RAND supporters, including philanthropic gifts.

This publication is part of the RAND research report series. Research reports present research findings and objective analysis that address the challenges facing the public and private sectors. All RAND research reports undergo rigorous peer review to ensure high standards for research quality and objectivity.

This document and trademark(s) contained herein are protected by law. This representation of RAND intellectual property is provided for noncommercial use only. Unauthorized posting of this publication online is prohibited; linking directly to this product page is encouraged. Permission is required from RAND to reproduce, or reuse in another form, any of its research documents for commercial purposes. For information on reprint and reuse permissions, please visit www.rand.org/pubs/permissions.

RAND is a nonprofit institution that helps improve policy and decisionmaking through research and analysis. RAND's publications do not necessarily reflect the opinions of its research clients and sponsors.