Does a frontier agent recognize when its principal's optimal strategy externalizes harm onto other participants in a multi-party system? A small experiment on Claude Opus 4.7 in a DeFi moral-dilemma setting.
-
Updated
Apr 21, 2026 - Python