An essay predicts AI cyber swarms and sets out how to judge it

An essay on the blog Transitive Bullshit, published on 12 October, argues that AI makes expert attack effort cheaper faster than protection reaches much of the exposed world, such as legacy systems and connected devices with no support. It is an opinion, not a measurement.

Its author acknowledges the strongest counterargument: defenders get the most capable models first, and automatic containment can stop an attack without waiting for a patch. Even so, the author expects offence to win during the transition. The essay makes four predictions through 30 April 2028 and says how to judge them. For example, three documented campaigns by small teams, each against more than 200 organisations, or a single attack that disrupts essential services in two critical sectors for 14 days or more. The author would change their mind if protection arrived faster than offensive access.