Shadow Work and Protocols: A Contingent Case for Negligence in Agentic AI
Calo and Paternolli make the case for negligence over strict liability in agentic AI, and put the protocol layer where liability should attach.
Ryan Calo and Veronica Paternolli connect protocol design to liability doctrine, and it's a bridge I haven't seen before.
Their contingent case is for negligence over strict liability for agentic AI. It rests on two things: the social utility of agents clearing away bureaucratic "shadow work," and the protocol layer, MCP and its kin, as the place where liability incentives should attach.
TLDR; the design of the protocol is a liability decision. When negligence doctrine reaches agentic AI, the choices baked into MCP will be exhibit A.
Shadow Work and Protocols: A Contingent Case for Negligence in Agentic AI
A contingent case for applying negligence rather than strict liability to agentic AI, with the protocol layer as the point where liability incentives should attach.