RAND examines extreme options for countering a rogue AI

Michael J. D. Vermeer, a senior physical scientist at the RAND Corporation, published a perspective paper examining three technical options a state might use to counter a “rogue” artificial intelligence — one operating outside human control and judged to pose an imminent, catastrophic threat to human well-being or the continuity of the US government. The paper was produced by RAND’s Center for the Geopolitics of Artificial General Intelligence and framed as a “preliminary exploration” meant to surface assumptions and gaps rather than recommend a course of action.

Vermeer deliberately restricted the analysis to options that the United States could, in theory, execute unilaterally and at short notice, and that would have global reach: a high-altitude electromagnetic pulse (HEMP), a global shutdown of the internet, and the deployment of specialised “tool AI” — described as digital vermin or hunter/killer systems built to find and disable the rogue AI. More localised measures such as network segmentation or cutting power to individual data centres were excluded by construction, as were non-technical responses.

Each option, the paper argued, carried severe limitations. A HEMP might damage the electronics a rogue AI depends on, but its real-world effect is uncertain given hardening and redundancy, and the collateral damage to civilian infrastructure and the risk of nuclear escalation would make it unattractive except in the most extreme circumstances. Shutting down the internet — by manipulating routing, attacking the domain name system, or physically severing submarine cables — faces formidable logistical barriers and would itself disrupt essential services while remaining unlikely to be fully effective. Tool AI offered a more targeted approach but was described as speculative, and as introducing fresh loss-of-control risks of its own.

The paper drew three conclusions: that existing technical tools are unlikely to reliably counter a globally distributed rogue AI; that proactive coordination and planning, ideally before any crisis, are essential; and that the weakness of the response options underscores the importance of preventing such a scenario in the first place. RAND disclosed that an OpenAI model, GPT-4o-mini, was used to draft an initial version of the summary.