In multi-agent systems, a valid conclusion can gain unintended authority as it moves downstream. That makes agent handoffs a security boundary of their own.