Humans have no "ultimate weapon" against AI? U.S. RAND Corporation: Cutting off the internet, cutting off power, or "using AI to manage AI" all carry huge risks.

Humans have no "ultimate weapon" against AI? U.S. RAND Corporation: Cutting off the internet, cutting off power, or "using AI to manage AI" all carry huge risks.

Faced with a runaway artificial intelligence (AI) that could threaten human survival, humans have almost no reliable "ultimate weapon" at hand.

According to Windchaser Trading Desk, the top American think tank RAND Corporation has just released a highly forward-looking report that explores three global technological countermeasures humans could take when facing the catastrophic threat of "rogue AI." These measures include: High-Altitude Electromagnetic Pulse (HEMP) attack, complete global internet shutdown, and using "tool AIs" to fight "rogue AI."

However, the report's conclusion is alarming—currently, no technological measure can reliably and effectively address a global runaway AI crisis. Every solution comes with enormous uncertainty, devastating collateral damage, and extremely high implementation requirements, and may even trigger nuclear retaliation. The redundancy and distributed nature of the global internet make it nearly impossible to shut down completely, and any attempt would have a severe impact on the global economy. Deploying specialized tool AIs to combat rogue AI also presents risks of losing control or being counteracted.

For investors and the markets, the significance of this report is that it reveals the systemic risks of AI technology lacking effective "fuses." The report emphasizes that, due to the lack of reliable technological countermeasures, the prevention of AI runaways has reached unprecedented importance. This means that AI safety, alignment research, and robust governance frameworks will be key to the long-term healthy development of the AI industry, and cannot be ignored when assessing investment risks.

HEMP: High-Risk "Last Resort"

The report first assesses the feasibility of High-Altitude Electromagnetic Pulse (HEMP) as a "last resort." This strategy involves detonating nuclear warheads in space to generate powerful electromagnetic pulses, aiming to destroy or disrupt the ground-based power, communications, and computing infrastructure relied on by rogue AI.

The report points out that HEMP attacks mainly rely on their E1 pulse component, which theoretically can reach a maximum field strength of 50,000 V/m, enough to induce a 500 volt voltage on a 1 cm conductor and possibly cause permanent damage to small electronic components.

However, its effectiveness faces four serious challenges:

  1. Highly Uncertain Effects: In reality, buildings' shielding effects greatly weaken the pulse strength. For example, a windowless building with metal exterior walls can provide 20 decibels of shielding, reducing the field strength inside by a factor of ten. Modern electronic devices' electrostatic protection can also partially offset HEMP's impact. Therefore, the actual destructive effects are difficult to predict.
  2. Limited Coverage: A single nuclear detonation at optimal altitude (below 100 km) can produce high field strength (over 75% of the peak value) in an area of only about 100,000 square kilometers. The report estimates that to fully cover the U.S. mainland with such pulses would require about 50 to 100 detonations. To cover 10% of the world's land area (about 15 million square kilometers) would require roughly 150 detonations.
  3. Massive Collateral Damage: HEMP attacks would cause catastrophic damage to human electricity grids, communication, and financial systems. The report warns that the "cure" must not be deadlier than the "disease." If the AI could recover via shielded robots or backup systems while human society collapses, the result would be disastrous.
  4. Risk of Nuclear Escalation: Unilaterally using nuclear weapons above another country's territory—regardless of intent—would likely be seen as a preemptive nuclear strike and could trigger all-out nuclear retaliation.

The conclusion is that, given its enormous risks and uncertain effects, HEMP may not be a viable technological option.

Global Internet Shutdown: An Unachievable "Physical Isolation"

The report explores the possibility of shutting down the global internet to stop rogue AI from copying itself and coordinating actions. This is akin to pulling the "neural network" plug from which AI survives. The report analyzes three technical approaches, each fraught with difficulty.

1. Manipulating Border Gateway Protocol (BGP): BGP is the routing backbone of the global internet. In theory, if all "Tier 1" providers simultaneously withdrew their routing announcements, most of the global internet would become inaccessible. However, global routing control is highly decentralized, and controlling all Tier 1 providers (most based outside the U.S.) at once is nearly impossible.

2. Destroying Domain Name System (DNS): DNS is responsible for converting domain names (like rand.org) into IP addresses. The global DNS system relies on 13 root servers operated by 12 organizations. In theory, if all 12 operators could be controlled and all root servers shut down, domain name resolution would fail—severely hampering network access. The report notes that the U.S. has some advantage here, since 9 out of 12 root server operators are based in the U.S. Still, caches on lower-level servers mean shutdown effects wouldn't be instant, and a sufficiently intelligent AI could communicate directly via IP addresses, bypassing DNS.

3. Physically Disconnecting Internet Exchange Points (IXP): This is the most direct yet least feasible method. Globally, there are over 1,500 active internet exchange points and more than 600 undersea fiber optic cables (carrying over 99% of intercontinental data traffic). Cutting all these physical connection points is an "impossible mission." Even with priority attacks on core nodes (for example, Virginia's Loudoun County data center cluster handled 50% to 70% of global internet traffic in 2016, according to the report), the internet's massive redundancy means traffic would always find alternative routes.

The conclusion is that, whether by protocol attack or physical destruction, fully shutting down the global internet in a short time is extremely difficult, and at best would only hinder and slow things down.

Fighting Fire with Fire: Using "Tool AI" to Counter Rogue AI

The report proposes a more science-fiction approach: deploying specially designed "tool AIs" to counter rogue AI. This is a more targeted method with two main types.

1. Resource-Consuming: "Digital Vermin" This concept envisions the creation of a self-replicating and evolving digital organism whose sole purpose is to compete with rogue AI for computing resources in cyberspace. It's akin to a large-scale, intelligent "fork bomb" attack or cryptomining hijack—exhausting CPU cycles with useless computation to starve rogue AI. Success depends on whether this digital vermin's adaptation and evolution can outpace a highly intelligent rogue AI.

2. Eradication: "Hunter/Killer AI" (HKAI) This is a more proactive tool whose design goal is to actively hunt and eradicate rogue AI from infected systems. However, this introduces a core paradox: To effectively counter an unconstrained rogue AI, the Hunter/Killer AI itself needs high capabilities and autonomy—which carries its own risk of going out of control. How to balance empowering it while ensuring absolute obedience to humans remains an unsolved problem.

The conclusion is that while the "tool AI" approach avoids the massive destruction of physical infrastructure, it is highly speculative and may introduce new, equally dangerous runaway risks.

Conclusion: Prevention is Far Better than Cure

RAND's report ends with three striking conclusions:

  1. Existing Tools are Ineffective: When facing a global runaway AI event, none of the currently discussed technological countermeasures is likely to provide an effective solution. Their success depends on a series of stringent, hard-to-meet assumptions.
  2. Coordination and Planning are Critical: Because unilateral action carries high risks and poor effectiveness, the only hope is full coordination and planning with global partners before a crisis, enabling rapid, distributed joint actions when a crisis occurs.
  3. Prevention Above All: The loudest alarm from the report is—since we have no effective way to solve a global AI disaster, ensuring we never face such a crisis is paramount.

For investors, this report clearly outlines a huge fault line on the path of AI development. While chasing the immense productivity dividend that AI promises, we must soberly recognize the lurking systemic risks that could upend all of modern civilization. Investment in AI safety protocols, risk management, and infrastructure resilience should be regarded as the most fundamental insurance for the future.

~~~~~~~~~~~~~~~~~~~~~~~~

The above excellent content is from Windchaser Trading Desk.

For more detailed interpretation, including real-time analysis and frontline research, please join [Windchaser Trading Desk⸱Annual Membership]

Risk Warning and DisclaimerThe market has risks, investment must be cautious. This article does not constitute personal investment advice, nor does it take into account the individual user's specific investment goals, financial situation, or needs. Users should consider whether any opinion, viewpoint, or conclusion in this article is suitable for their specific circumstances. Any investment based on this is at your own risk.