Tag
The article argues that the key question in AI agent safety should shift from preventing misbehavior to proving what agents did and were authorized to do in multi-agent systems, highlighting an infrastructure gap, particularly in finance.