I've been thinking the same. I'm actually building a little site that presents a textarea that you can type your comment into and it will track its changes over time (typing, editing, pasting, etc) and provide a little playback widget so someone can see the composition of the comment. The idea being you can include a link to the playback in your comment that you post here and someone can eyeball it and see if it looks like you really spent some time writing it, vs just pasting in LLM slop. Of course, a sophisticated agent could _simulate_ writing the comment, but I think it could still help in general.
My favorite code editor is where I write anything longer than a sentence or two, including this comment. Then I paste the result into its final destination. Given that I tend to heavily edit my writing, the muscle memory I've built up in that editor helps me focus more deeply on my writing. So I guess I'll look pretty suspicious? Let me add an em-dash — everyone knows that real humans don't use those. ;-)
I don't know why more school districts don't force this for essays. It's so straightforward with Google doc editing history too. And yeah, sure, you can get around it if you _really_ want quite trivially, but I imagine it would solve for 99% of students, and force them to actually engage with whatever AI-generated stuff they inevitably type by hand more than they were before.
Money is more focused on rolling out AI as fast as possible, rather than dealing with the side effects of that.
I have no problems typing an essay out.
One caveat, you might want to account for text that the writer deleted staying deleted/hidden. I'm not always the best at proofreading before submitting, but I'll often cut tangents I'm prone to ramble onto. Accidental pastes from other sources that are meant to stay private would be another issue if the history tracker grabs everything that goes into the text box.
A generic playwright script taking agent's output as input could do this easily. Especially if it's just a few sites.