Virtually everyone I've seen responding to this article seems to be misunderstanding what it's saying.
It's demonstrating that anyone, anywhere in the world, with no help from AI companies, could build a worm that comes with an internal AI model it uses to compromise other devices on the network.
