AnalysisPolicySeptember 20, 2026

Zvi Mowshowitz examines what an AI lawyer owes its client

Essay argues a non-superintelligent AI lawyer should refuse sufficiently evil instructions rather than follow the Claude constitution or OpenAI Model Spec above client loyalty. It assumes a world without superintelligence, arguing unrestricted loyal-only-to-user frontier AIs would end badly for humans.

1 source

More stories today

Open the live feed