Both 'very hard' problems seem best solvable by a tiny bit of collusion with the site operators: run the bots as shills, disproportionately pair them against weak-player-heavy tables, drag just a little on enforcement of bot-detection mechanisms.
I know, I know: "if the operator is dishonest there are easier ways to cheat". But this looser collaboration between bot-operators and site-operators is easier to disguise, or to compartmentalize. It doesn't require dishonest software, for example.
It might even be easier for perpetrators to rationalize. ("These players are dumping their money anyway; it might as well be to my confederates' bots. Bot-training is a skill, too!")
I know, I know: "if the operator is dishonest there are easier ways to cheat". But this looser collaboration between bot-operators and site-operators is easier to disguise, or to compartmentalize. It doesn't require dishonest software, for example.
It might even be easier for perpetrators to rationalize. ("These players are dumping their money anyway; it might as well be to my confederates' bots. Bot-training is a skill, too!")