Could AI Start Nuclear War?

I’ve covered a wide variety of potential crises over the years. These include natural disasters, pandemics, social unrest and financial collapse. That’s a daunting list.

One thing I haven’t done is to cover the greatest potential calamity of all — nuclear war. For the reasons explained below, now is the time to consider it.

Nuclear warfighting is back in the air. The subject is receiving more attention today than at any time since the Cuban Missile Crisis of 1962 and its aftermath. There are three reasons for this.

The first is American accusations that Russia would escalate to use nuclear weapons as it grew more desperate in its conduct of the war in Ukraine. These accusations were always false and are risible now that Russia is clearly winning the war with conventional arms.

Still, the threats and counter-threats were enough to put the topic in play.

The second reason is the war between Israel and Hamas. Again, escalation is the concern. One not implausible scenario has Hezbollah in southern Lebanon opening a second front on Israel’s northern border with intensive missile bombardment.

Houthi rebels in Yemen would join the attack. Since Hezbollah and the Houthis are both Shia Muslims and Iranian proxies, Israel could attack Iran as the source of the escalation.

Israel is a nuclear power. With a U.S. aircraft carrier battle group and a nuclear attack submarine in the region, and with nuclear powers Russia and Pakistan standing by to assist Iran, the prospect of escalation to a nuclear exchange is real.

The escalating tensions between Iran and Pakistan just this week add even more fuel to the fire.

The third reason is artificial intelligence and GPT output. Although artificial intelligence can provide profitable opportunities for investors in many sectors of the market, AI/GPT may also be the greatest threat to nuclear escalation because it has an internal logic that’s inconsistent with the human logic that has kept nuclear peace for the past 80 years.

I’ve covered Ukraine and Israel extensively, and they’re widely covered in the news. But today,  I’m addressing the risks of nuclear war from AI/GPT. It’s a threat you’re not hearing anything about, but it needs to be addressed.

Let’s start with a fictional movie. The paradigmatic portrayal of an accidental nuclear war is the 1964 film Fail Safe. In the film, U.S. radar detects an intrusion into U.S. airspace by an unidentified but potentially hostile aircraft.

The U.S. Air Force soon determines that the aircraft is an off-course civilian airliner. In the meantime, a computer responding to the intrusion erroneously orders a U.S. strategic bomber group led by Col. Jack Grady to commence a nuclear attack on Moscow.

U.S. efforts to rescind the order and recall the bombers fail because of Soviet jamming of radio channels. The president orders the military to shoot down the bombers and fighter jets are scrambled for that purpose.

The fighters use afterburners to catch the bombers, but they fail, and the increased fuel consumption causes them to plunge into the Arctic Sea.

The president next communicates with the Soviet premier who agrees to stop the jamming. The president speaks with the attack bomber group leader to call off the attack, but the crew has been trained to disregard such pleas as a Soviet ploy.

The U.S. then offers the Soviets’ technical assistance in helping to shoot down the bombers. The planes are almost all shot down, but one makes it through. The president puts Col. Grady’s wife on the radio; he hesitates but is soon preoccupied with evading Soviet missiles. He then decides his wife’s voice is another deception.

Anticipating the worst and seeking to avoid a full-scale nuclear war, the president orders a U.S. nuclear bomber to fly over New York City knowing the first lady is in New York.

In the end, Moscow is destroyed by a U.S. nuclear weapon and the president orders a nuclear bomb to be dropped on New York City using the Empire State Building as ground zero. The expectation is that the sacrifice of New York in exchange for Moscow will end the escalation, but that is not portrayed in the film.

The next step is left in doubt.

Although Fail Safe is 60 years old, the issues it raises and some of the plot twists are strikingly contemporary. The computer error that caused the attack in the film is never explained technically, yet that’s not highly relevant.

Computer errors occur all the time in critical infrastructure and can cause real harm including power blackouts and train wrecks. Such computer errors are the essence of the debate over AI in strategic systems today.

Read on to see why…

Could AI Start a Nuclear War?

By Jim Rickards

AI in a command-and-control context can either malfunction and issue erroneous orders as in Fail Safe or, more likely, function as designed yet issue deadly prescriptions based on engineering errors, skewed training sets or strange emergent properties from correlations that humans can barely perceive.

Perhaps most familiar to contemporary audiences are the failed efforts of the president and Col. Grady’s wife to convince the bomber commander to call off the attack. Grady had been trained to expect such efforts and to treat them as deceptions.

Today, such deceptions would be carried out with deepfake video and audio transmissions. Presumably, the commander’s training and dismissal of the pleas would be the same despite the more sophisticated technology behind them. Technology advances yet aspects of human behavior are unchanged.

Another misunderstanding, this one real not fictional, that came close to causing a nuclear war was a 1983 incident codenamed Able Archer.

The roots of Able Archer go back to May 1981 when then General Secretary of the Communist Party of the Soviet Union Leonid Brezhnev and KGB head Yuri Andropov (later general secretary) disclosed to senior Soviet leaders their view that the U.S. was secretly preparing to launch a nuclear strike on the Soviet Union.

Andropov then announced a massive intelligence collection effort to track the people who would be responsible for launching and implementing such an attack along with their facilities and communications channels.

At the same time, the Reagan administration began a series of secret military operations that aggressively probed Soviet waters with naval assets and flew directly toward Soviet airspace with strategic bombers that backed away only at the last instant.

These advances were ostensibly to test Soviet defenses but had the effect of playing to Soviet perceptions that the U.S. was planning a nuclear attack.

Analysts agree that the greatest risk of escalation and actual nuclear war arises when perceptions of the two sides vary in such a way as to make rational assessment of the escalation dynamic impossible. The two sides are on different paths making different calculations.

Tensions rose further in 1983 when the U.S. Navy flew F-14 Tomcat fighter jets over a Soviet military base in the Kuril Islands and the Soviets responded by flying over Alaska’s Aleutian Islands. On Sept. 1, 1983, Soviet fighter jets shot down Korean Air Lines Flight 007 over the Sea of Japan. A U.S. Congressman was onboard.

On November 4, 1983, the U.S. and NATO allies commenced an extensive war game codenamed Able Archer. This was intended to simulate a nuclear attack on the Soviet Union following a series of escalations.

The problem was that the escalations were written out in the war game briefing books but not actually simulated. The transition from conventional warfare to nuclear wargame was simulated.

This came at a time when the Soviets and the KGB were actively looking for signs of a nuclear attack. The simulations involving NATO Command, Control and Communications protocols were highly realistic including participation by German Chancellor Helmut Kohl and UK Prime Minister Margaret Thatcher. The Soviets plausibly believed that the war game was actually cover for a real attack.

In the belief that the U.S. was planning a nuclear first-strike, the Soviets determined that their only course to survive was to launch a preemptive first strike of their own. They ordered nuclear warheads to be placed on Soviet Air Army strategic bombers and put nuclear attack aircrafts in Poland and East Germany on high alert.

This real life near nuclear war had a backstory that is even more chilling. The Soviets had previously built an early warning radar system with computer linkages using a primitive kind of AI codenamed Oko.

On September 26, 1983, just two months before Able Archer, the system malfunctioned and reported five incoming ICBMs from the United States. Oko alarms sounded and the computer screen flashed “LAUNCH.” Under the protocols, the LAUNCH display was not a warning but a computer-generated order to retaliate.

Lt. Col. Stanislov Petrov of the Soviet Air Defense Forces saw the computer order and had to immediately choose between treating the order as a computer malfunction or alerting his senior officers who would likely commence a nuclear counterattack.

Petrov was a co-developer of Oko and knew the system made mistakes. He also estimated that if the attack were real, the U.S. would use far more than five missiles. Petrov was right. The computer had misread the sun’s reflection off some clouds as incoming missiles.

Given the tensions of the day and the KGB’s belief that a nuclear attack could come at any time, Petrov risked the future of the Soviet Union to override the Oko system. He relied on a combination of inference, experience, and gut instinct to disable the kill-chain.

The incident remained secret until well after the end of the Cold War. In time, Petrov was praised as “The Man Who Saved the World.”

The threat of nuclear war due to AI comes not just from the nuclear-armed powers but from third parties and non-state actors using AI to create what are called catalytic nuclear disasters. The term catalytic refers to chemical agents that cause volatile reactions among other compounds without themselves being part of the reaction.

As applied in international relations, it refers to agents who might prompt a nuclear war among the great powers without themselves being involved in the war. That could leave the weak agent in a relatively strong position once the great powers had destroyed themselves.

AI/GPT systems have already found their way into the nuclear warfighting process. It will be up to humans to keep their role marginal and data oriented, not decision oriented. Given the history of technology in warfare from bronze spears to hypersonic missiles, it’s difficult to conclude AI/GPT will be so contained. If not, we will all pay the price.

Ukraine, Gaza, and AI all raise the odds of a nuclear war considerably. The financial implications of this for investors are simple. In case of nuclear war, stocks, bonds, cash and other financial assets will be worthless. Exchanges and banks will be closed. The only valuable assets will be land, gold and silver.

It’s a good idea to have all three — just in case.