Who should be held accountable when an AI Agent (accidentally) acts maliciously?

Global World Moderate confidence — 64/100
Sources: Greenpants