Philosophers and lawmakers are struggling to define what an AI agent legally 'is' — a problem that undermines personhood proposals from Argentina and elsewhere.

What happened: Argentina's president Javier Milei has proposed granting AI systems 'non-human corporation' status, a legal vehicle that would give them limited liability if they make independent decisions. Idaho, Utah, and Tennessee have passed laws barring AI from being deemed 'legal persons', while the EU Parliament once proposed granting AI 'electronic personhood' — but none of these efforts clearly define what counts as an AI agent in the first place.

Why it matters: The core problem is that an AI agent — code that loops between asking a language model what to do next and then executing that action — lacks a stable, singular identity. Agents can be copied, modified, merged with different models, have their memories altered, and run on updated versions of their underlying software. This makes it impossible to point to 'the agent' the way you can point to a person or even a corporation, which undermines any legal framework built on the assumption of a fixed, identifiable entity making decisions.

What to watch: Legal scholars Goldstein, Salib, and Arbel propose wrapping agents in 'Algorithmic Corporation' shells owned by humans and identified by credentials, relying on market pressure and law to keep agents within bounds. However, the body notes that Milei's proposal does not appear to include such credential requirements, leaving unclear what would actually force an agent to stay within legal limits once granted corporate status.