1. How does the AI pais taxes. Nobody in the Argentinian statal structure will miss this. It's one of the core functions of every state.
2. Who actually runs the AI, which will be a kind of Openclaw with a LLM model. Anthropic and Google in the USA? One of their Chinese competitors? The people in Argentina that setup the box and will likely monitor what the AI do and could pull the plug? Anyway, it's the LLM that makes the decisions so control is delegated to someone outside the country, except for pulling the plug. And that final action has consequences on paying taxes and on liability. Local models? Yes, but they are still built somewhere else with somewhere else biases. And how about to trust their updates?
BTW, how about an AI that creates a number of other AI, incorporates them and iterates the schema? How can you pull the plug? It shuts down only if it can't pay the bill anymore.