Should every AI agent have a unique ID? A case for proportional identification based on risk, liability, privacy and audit.