Not really interested as I'm working on physics field now, but did my Bachelor on scam tracing in Ethereum. At that time I used mathematical methods and I think every 5th transaction led to potential money laundering cases. AI can easily separate those, so I think that might be an interesting theme for paper to you. You could probably freeze transactions before it reaches distribution point. Most of those transactions are victims --> pyramid or some other scheme --> covering chain [A --> B--> C -->D] --> laundering point such like currency conversion point (ETH to BTC or some other) --> scammer (at this point you probably cannot trace them as exchange markets have many users converting at the same time. Start and end points are easy to noticed based on number of in and out transactions. Meanwhile, those chains are often 1-to-1 transaction. Also, pyramid schemes will probably not be going to last more than few days.
But if you are studying criminology, you probably know more about this topic than me. Just pointing that AI would easily identify these cases as long as you would provide enough data to train model. I'm almost sure even ChatGPT would do that with the description I just provided :D