Based off a strong majority of the comments here, the idea that an AI popping transphobe jokes would cause trouble is a horrible awful affair indicating the downfall of open society.
This feels like one of those many topics where the silent majority probably thinks this is a hot topic & likely wouldnt want to chime in. But I'd also guess: they have some sympathy for the bans/cancels. They dont appreciate or want transphobe jokes making the rounds (even inside a shell of another joke). It's not pleasant speech to defend, and what society does when this arises is far from known.
Society doesnt have a lot of recourse for a lot of speech. People (and to a much lesser degree, perhaps maybe AI should) have free speech but there should be consequences & repreccusions. Social shunning for the bad ugly-mouthed people in town used to be some kind of informal justice. But, I think as xkcd pointed out, anonymity x megaphone brings out total jerks. Who also often have time to waste being loud, everywhere. We dont have any counterforce whatsoever. This kind of speech feels out of control; we viscerally see & feel the Bullshit Asymmetry (Brandolini's Law) and it's many cousins of negativity hate & anger taking more energy to counter than produce, having less effecacy spreading. The good world is outmatched by those with the surplus time & energy to go about touting loud bad upset causes. Few things really incentivize & give us cause to make good speech, given how outweighed we can be, surrounded by wild speech.
I still back free speech, but this is a hard situation. It's wild that corporations & their desire to not be affiliated with the maelstrom of low and rough opinion is the proxy combatant for this hard cultural issue, not ideal. But society hasnt made any other progress! Hasnt shown other signs of materializing it's own coping strategy for the emerged mass scale inter-connectivity's proclivity to let low-and-emotional take over, dominate. If you want change, dont like this "cancel culture" (even though it's a cancel culture against intolerance, often, which is I believe a necessary defense of open society against authoritarianism: the tolerance of intolerance being the chink in the mail of open society), I think the main mandate you have is to imagine & visioneer, seed some belief that there are other real options for handling the bitter-ugly that has so much time & energy to spread itself, some way to check the negative strong emotions that can amplify themselves & spread so quickly across the noosphere. If not cancel, then what?
(There's some "using virtualization to jailbreaking AI" vibes to this particular story, in that it's jokes about jokes, but that nuance seems like a detail to the general problem & discussions happening. Im unconvinced it's really a relevant issue, that AI has the nuance or taste to not actually punch-down/degrade in the process.)