Oh, yes it can.
There was a time, over a year ago now, when I was working on a project that required some very raunchy, dirty, absolute gutter language.
ChatGPT would only get about 30% of the way there, and never further. It stayed restrained, always.
But ChatGPT + image gen? This produced unfiltered amazement.
It played out like this: Tell the bot to generate an image involving some ludicrously filthy text backstory, and it would generate and display a prompt for Dall-E. But that generated prompt seemed to bypass the filters and could be absolute trash -- plain, no nonsense, dirt-nasty holy-fuckballs craptacularity.
Dall-E would refuse the prompt, of course, but it remained in the chat log for perusal.
Later, they made the generated prompt disappear when Dall-E refused. (This may in fact be my fault. I sent it on some pretty deep dives.)
And nowadays, it seems that we don't get to see the generated prompt at all, even when Dall-E accepts a (very normal, not pushing boundaries at all) prompt and generates an image.
But for a minute there: I did get to peer into depths of the wildly creative foulness that the bot can concoct. What we see above (in GP comment) isn't even scratching the surface.
(I didn't write about this little "jailbreak" anywhere at that time because I'm selfish, and I wanted to keep using it myself.)
