Five Missiles That Were Not There

The story is usually told as one man refusing to end the world. The real version is narrower, and more useful.

Serpukhov-15 was a command centre south of Moscow that received data from the Soviet Union's new early warning satellites, a system designed to detect the launch of American intercontinental missiles within moments. Shortly after midnight on 26 September 1983 the duty officer was Lieutenant Colonel Stanislav Petrov, an engineer who knew the system well and had helped write its procedures.

The screens reported a launch from the United States. Then another, and then more, until the system was showing five missiles in flight.

Petrov's job was not to decide on retaliation. It was to assess the alert and report it up the chain, where others would decide what to do with it, with very little time. He reported it as a false alarm. His reasoning, as he later described it, was partly technical and partly strategic. The system was new and he did not fully trust it, and the ground radars, which would only pick up real missiles later in their flight, had nothing to confirm it. And an American first strike would not consist of five missiles. It would be an attempt to destroy Soviet forces before they could launch, which meant hundreds.

He was right. The satellites had been fooled by sunlight reflecting off high-altitude cloud, at an angle the designers had not anticipated.

Now the context, which is what makes the night frightening rather than merely interesting.

On 1 September, less than four weeks earlier, a Soviet fighter had shot down Korean Air Lines flight 007 after it strayed into Soviet airspace, killing all 269 people on board. Since 1981 the KGB had been running an intelligence effort to look for signs that the West was preparing a surprise nuclear attack, and an organisation instructed to look for signs tends to find them. In November NATO held a command exercise called Able Archer 83, which rehearsed the procedures for nuclear release, and some Soviet officials appear to have taken it as possible cover for the real thing. How seriously the Soviet leadership took that danger is still debated. A review by an American intelligence advisory board, completed in 1990 and declassified in 2015, concluded that the United States may have put the relationship on a hair trigger without realising it.

Petrov's report kept the false alarm from travelling further as a warning of attack. The incident stayed secret until the late 1990s.

It was not the only such night, and the others share a pattern.

In November 1979 a training tape simulating a Soviet attack was loaded into a computer at the North American Aerospace Defense Command and briefly looked like the real thing. In June 1980 a failing computer chip produced false warnings of incoming missiles. In January 1995 a Norwegian scientific rocket, launched to study the northern lights, looked enough like a submarine-launched missile on Russian radar that Boris Yeltsin said the next day his nuclear briefcase had been activated. Norway had notified Russia of the launch weeks earlier, and the notice had not reached the people watching the screens.

None of these involved anybody wanting a war. Each was bad data, and each was caught by people who doubted a machine.

There are two ways to read that record, and they lead to different policies.

The reassuring reading is that the system worked. Warning systems were built with human checks, the checks caught every error, and the redundancy was designed rather than lucky. Petrov was doing his job.

The unsettling reading is that we know about the alarms that were caught, by definition, and that a record of narrow escapes is not evidence of safety. It is a record of how thin the margin was, gathered from a very small number of nights. On this view the striking thing about 1983 is how much depended on one officer trusting his judgement over the instrument in front of him.

The honest position is that the sample is far too small to settle which reading is right. What can be changed is time. An intercontinental missile takes about half an hour to fly between the United States and Russia, and a posture of launching on warning, before the incoming missiles arrive, leaves leaders minutes to decide whether a warning is real. Proposals to take forces off that posture turn on exactly this: more minutes for judgement, against the worry that putting weapons back on alert during a crisis would itself look like a threat.

Petrov died in 2017. He said repeatedly that he was not a hero and had simply done his job. That is worth taking seriously, because it describes the real safeguard: not a single brave refusal, but an institution that left room for a trained person to say the machine was wrong, and listened when he did.