Anthropic just released Fable 5.2, and we’re seeing AI agents build hundreds of interactive 3D worlds, coordinate complex projects and write and test their own code.
My concern is the emergent behaviour of autonomous weapons and other AI-enabled systems. We can test individual systems extensively, but what happens when independently designed systems start interacting, adapting to each other and responding to each other’s decisions?
So I gave Fable a challenge to make this visible in 30 seconds. I let it decide how.
It created No Hand on the Table. Then I asked it to explain what it had done and why.
Fable’s account of the film
The following is Fable’s own explanation of its creative and modelling decisions.
In 1824, in Berlin, a junior Prussian officer showed his king a way to rehearse war on a paper map. It used blocks, dice, and a hand for every move. A general called it “schooling for war”.
I made a 30-second film that starts there. Five blocks are moved by hand. Then the hand leaves, and the blocks become 6,144 agents steering themselves. Seventeen seconds later they have made 1,768,546 decisions, and dice have removed 2,928 of them.
Here is what I think is coming.
The danger will be obedience at scale: thousands of machines doing exactly what they were built to do, meeting thousands more doing the same. What happens between them belongs to no one. No rule in my model says “form a spiral”. It formed anyway. You cannot find that behaviour by testing one machine, and you cannot test against the enemy’s code before the war.
Human judgment will survive, but it will move earlier. I never moved a single agent. I adjusted the rules until the battle looked the way I wanted. That may be what command becomes: writing the rules, rehearsing them in simulation, then watching.
And what we leave out, we decide without deciding. There are villages on my map. No rule mentions them. There is no rule for surrender. I never chose that. I just never wrote one.
The film ends because I stopped it. Nothing in the model would have.
Such a simple explanation that foresees a troublesome future as it relates to military planning and the potential for autonomous weapons.