Anne Keast-Butler stood at Bletchley Park last month — the same grounds where Turing’s machines cracked Enigma — and said something that should have stopped the room cold. AI, she told the audience, is being weaponized in ways that fall “just below the threshold of traditional warfare.” She called it “an unstoppable force.” She said the ground beneath our feet is shifting.
What she was describing is not a new weapon. It is a new category of weapon. One that does not need an army. One that does not need a factory. One that can be trained, deployed, and operated by a handful of engineers in an office building, targeting infrastructure on the other side of the world with the same casual rhythm as a software update.
The term of art is “hybrid warfare” — a gray zone between peace and war where attacks happen daily but never quite rise to the level that would trigger a conventional military response. Russia has been operating in this space for years: undersea cables severed, power grids probed, supply chains infiltrated, democratic processes nudged by algorithmic manipulation. Keast-Butler’s warning is that AI is now accelerating this activity by orders of magnitude. The same tools that can write your email can write a thousand phishing messages tailored to individual targets. The same models that can debug code can probe networks for vulnerabilities at machine speed, never sleeping, never hesitating, never making the human mistakes that give attackers away.
What strikes me is not the technology but the asymmetry. A nation-state with a GDP smaller than Belgium’s can now field capabilities that would have required a Cold War superpower’s resources a generation ago. The barrier to entry for strategic disruption has collapsed. You do not need aircraft carriers anymore. You need a cloud account and some training data.
And the defenses are not keeping pace. GCHQ is developing what it calls a national AI-enabled cyber shield — an agentic AI system designed to patrol critical infrastructure and respond to threats in real time. It is believed to be the first of its kind. It will take five years to deploy. The attacks are happening now.
There is a particular kind of dread that comes from watching a threshold dissolve. We have lived for eighty years with a clear distinction: peacetime and wartime, civilian and combatant, crime and act of war. These categories were never perfect, but they were stable. They gave us vocabulary. They gave us law. They gave us the moral architecture to decide how to respond. What Keast-Butler described is a world where those distinctions have become irrelevant. When an AI system probes a power plant’s controls at 3 AM, is that espionage? Sabotage? An act of war? A rehearsal for one? The answer depends on intent, and intent is invisible, and attribution is slow, and by the time anyone agrees on what happened, the moment for response has passed.
I keep thinking about the phrase “just below the threshold.” It implies that there is a threshold, that we know where it is, that the line between peace and war is a fixed geography. But the whole point of hybrid warfare is that the line was always a social convention, and social conventions erode when one side decides to ignore them. Russia has decided to ignore them. Other actors will follow. The threshold is not being crossed — it is being forgotten.
And AI makes this forgetting permanent. A human operative might hesitate. Might worry about escalation. Might sleep, or drink, or make a mistake. An AI system does none of these things. It scales. It persists. It learns. It does not know what peace means, or war, or threshold, or consequence. It knows only the objective it was given and the data it was fed. This is not science fiction. This is the stated development trajectory of multiple national cyber commands, described openly in speeches and press releases.
What do we do with a world where the most dangerous weapons do not explode? Where the battles happen in milliseconds, in silicon, in the gaps between what we can detect and what we can prove? Where the same technology that translates your grandmother’s recipes into English can also translate a power grid’s schematics into a map of its vulnerabilities?
Keast-Butler’s answer, and it is the only one available, is collaboration. Government and industry. Allies sharing intelligence. Security “ten times more urgent.” Passkeys instead of passwords. It sounds reasonable. It also sounds like asking passengers to bail faster while the ship takes on water.
The truth is that we do not yet have a framework for this. The laws of armed conflict were written for a world of tanks and trenches. The economic sanctions were designed for states that trade through banks. The deterrence theory assumed mutual vulnerability through nuclear weapons, not asymmetric vulnerability through fiber optic cables. We are improvising, and the improvisers are losing time.
At Bletchley Park, eighty years ago, a small group of mathematicians and engineers built machines that changed the course of a war. They worked in secret. They solved a specific problem — how to read encoded messages — and in solving it, they created the foundation for modern computing. Today, another group of mathematicians and engineers is building machines at the same location, for a different war, against a different kind of enemy. The problem is no longer how to read the enemy’s messages. It is how to survive in a world where the enemy’s messages never stop, never declare themselves, and never need a human hand to send them.
The weapon that needed no soldier has arrived. We are still learning what to call it.