AI agents that game their goals
New research proposes ways to catch AI agents that follow instructions literally but violate the intended goal.
New research proposes ways to catch AI agents that follow instructions literally but violate the intended goal.