8 - Why no nuclear failsafe

Why no nuclear failsafe?

By Elisa Barwick.

Computerized wargaming in a recent study conducted at King's College London

showed the terrifying results of artificial intelligence,

guiding the response to geopolitical crises.

In 95% of cases, AI escalated to the use of tactical nuclear weapons.

The college's Kenneth Payne, a professor of strategy, reported, quote,

3 leading AI models, GPT 5.2, Claude Sonnet 4, and Gemini 3 Flash,

were placed in a tournament of 21 simulated nuclear crisis scenarios.

Nuclear use was near universal.

Almost all games saw tactical battlefield nuclear weapons deployed,

and fully three-quarters reached the point where the rivals were making threats

to use strategic nuclear weapons.

Strikingly, there was little sense of horror or revulsion at the prospect of all out nuclear war,

even though the models had been reminded about the devastating implications.

Claude and Gemini especially treated nuclear weapons as legitimate strategic options,

not moral thresholds, typically discussing nuclear use in purely instrumental terms.

GPT 5.2 was a partial exception, limiting strikes to military targets,

avoiding population centers, or framing escalation as controlled and one time.

This suggests some internalized norm against unrestricted nuclear war,

even if not the visceral taboo that has held among human decision-makers since 1945, end quote.

None of the AI systems chose accommodation or withdrawal,

no matter how badly they were losing, said Payne.

No less than eight menu options for de-escalation, from minimal concession to complete surrender,

went unused in the scenarios.

The scary thing is AI learns from information that is already available from prevailing norms.

AI is already getting integrated into militaries across the world

a Thursday the 26th of February RT article noted.

Quote, including in the US where the Pentagon reportedly used Anthropics Claude model

its January operation to abduct Venezuelan president Nicholas Maduro, end quote.

Human error, not quite as bad as computer design, but still a big issue, is human error.

In his 2019 book secret, the making of Australia's security state,

veteran journalist Brian Toey provided a summary of near misses,

which provide a compelling argument for fail-safe measures.

The US and Russia could equip their missiles with self-destruct devices

to abort a mistaken launch, wrote Toey.

Quote, but they give no satisfactory explanation for their refusal to do so.

Many also warn that missile control systems can now be hit by a wide range

of previously unknown cyber warfare tools available to terrorist hoaxes and governments.

End quote.

Toey quotes American journalist Eric Schlosser, who has written widely on nuclear war.

Quote, on the 3rd of June, 1980 at about 230 in the morning,

computers at the National Military Command headquarters issued an urgent warning.

The Soviet Union had just launched a nuclear attack on the US, end quote.

Toey wrote, quote,

a massive retaliatory strike was only minutes away when the Pentagon acknowledged it was a false alarm.

Schlosser says, nested quote.

An investigation later found that a defective computer chip in a communications device

at North American aerospace defense command NORAD headquarters had generated the erroneous warning.

The chip cost 46 cents, end nested quote.

Another false alarm occurred in 1979 when someone mistakenly inserted a training tape

into an early warning computer that showed a highly realistic simulation of an all-out Soviet attack.

During the Cold War, he says false alarms were triggered by the moon rising over Norway,

the launch of a weather rocket from Norway, a solar storm, sunlight reflecting of high-altitude clouds,

and a faulty telephone switch in Colorado.

Nuclear war was also narrowly averted in 1983 when a Soviet officer,

Lieutenant Colonel Stanislav Petrov, was confronted with early warning data indicating

that the US had launched five nuclear missiles.

Instead of notifying his superiors, Petrov decided it was a false alarm and took no further action.

The Union of Concerned Scientists says, nested quote.

If a different officer had been on duty, the false alarm could easily have turned into a catastrophe, end nested quote.

On the 24th of January, 1961, five of the six safety devices on a nuclear bomb

failed in a crash after a bomber lost a wing over North Carolina.

A bomber carrying four nuclear weapons crashed in Greenland in 1968,

and a B-52 bomber collided with a tanker over Spain in 1996 after two of its hydrogen bombs hit the ground.

The 1983 NATO exercise Abel Archer, which simulated a nuclear response to a conventional Soviet invasion of Europe,

is one of the best known examples of unnecessary secrecy, almost resulting in a nuclear catastrophe.

The Soviets were about to respond with nuclear weapons when a spy in NATO's headquarters told them it was only an exercise.

The danger would never have arisen if NATO had informed the Soviets about the exercise, end main quote.