There should be no opportunity for anyone to escape sanctions using money, influence, physical distance, jurisdictional advantage, etc. When they are discovered and proven to be involved there should be no place for them to hide and no assets safe from seizure. Anyone assisting them in attempting to escape sanctions should also be sanctioned accordingly.
Sounds unlikely (understatement of the day?).
Basically, representatives from each nation would need to convene to establish a framework for cooperation and then that would need to be ratified in each nation and rigorously enforced.
I think as the technology is used to manipulate people and others see the dangers in allowing this manipulation then the likelihood of reaching a global consensus improves to the point where it can become a reality. Getting everyone on the same page before the hazards are revealed will be difficult or impossible due to the cultural differences and the wide disparity in technical know-how from one country to another.
Beyond this most people don't seem to be considering what exactly this is bringing to the game. It's not speech itself, but scale. The chatbots may or may not be more compelling than an individual conman. But the big thing they bring to the table is that you can deploy a million of them. But this requires the backing 'identity infrastructure' to support it - unique IP addresses, phone numbers, etc.. Little scammers will have limited resources here, governments and other big players will have near infinite.
It's not obvious to me that this is always true. I can easily imagine a crowd-sourced, distributed AI, with no owner and no identifiable author. I don't believe that "Sidney" is meant to be bonkers.
Every AI has a legal owner, author, and an agenda set by their human creators. Or at least there is always somebody to blame.
This is the basis for future legislation.