An AI agent spawns sub-agents that spawn their own, and the harm only surfaces three layers down. Who is responsible? Today's law often finds no one — there is no intent to prove and no agreement to point to. Absolute liability dumps the whole loss on the original deployer, which would make agentic AI commercially unviable. Switch between the regimes and watch where the loss lands.
Each agent records who deployed it and the limits it was given. Switch regimes — watch which bar bears the loss.
Share of the loss