October 1962. For thirteen days, Washington and Moscow stood on the brink of nuclear war, over missiles installed in Cuba.
What is most frightening, in hindsight, is not just the tension. It is the slowness. Messages between Kennedy and Khrushchev took hours to arrive, passing through embassies, translators, intermediaries. In the middle of a crisis where every minute counted, the two most powerful men in the world could not speak to each other directly.
The following year, a direct line was installed between the two capitals.
Sixty-three years later, the same idea is back. For artificial intelligence.
What was decided
At the end of a three-day summit in Washington between their leaders, China and the United States announced on Saturday that they had agreed to create a dedicated channel for AI-related incidents. They also agreed to speed up work on crisis communications between their militaries.
According to several American media outlets, a dialogue devoted to the most advanced AIs is also expected to open by November.
It is an interesting reversal. On Friday, we reported how AI bosses had gone to the UN to demand global rules, and how the American government had said no. No common rules, then. But a way to talk to each other when something goes wrong.
What it could be used for
Imagine that an AI developed in one of the two countries attacks computer systems in the other. Not on its government's orders: by accident, like the OpenAI agent that interacted with US government websites this month.
Seen from the other side, nothing distinguishes an accident from an attack. Public systems are affected, the origin appears foreign, the temptation to retaliate is immediate.
A direct line makes it possible to pick up and say: "It was us, it wasn't intentional, here is what happened." That is very little. And sometimes it is the whole difference between an incident and a crisis.
The line installed in 1963 between Washington and Moscow was neither a telephone nor red. It was a teleprinter: a machine that sent text, letter by letter. Voice had been ruled out on purpose. In writing, each side has time to translate correctly, to reread, to think before answering. A heated conversation, full of nuances and possible misunderstandings, seemed far too dangerous. The line later moved to fax, then, in 2008, to a secure computer link. The red handset on the president's desk is a invention of the cinema.
What it does not solve
A direct line does not prevent any incident. It only prevents an incident from being misunderstood.
It works only if both sides pick up, and tell the truth. It covers only two countries, whereas advanced models are also being developed elsewhere, and above all in private companies that do not go through diplomats.
And it arrives late: the incidents themselves have already begun.
What we take away
Throughout the Cold War, Washington and Moscow never agreed on the essentials. But they did agree on one thing: being able to call each other.
That is exactly what has just happened with AI. No common rules, no trust, but a thread between the two capitals. We would have liked more. We could have had much less.