If by "traditional solution" you mean a bunch of data is fed into creating an ML model and then your individual transaction is fed into that, and it spits out a fraud score, then no, they'd not using LLMs, but at this high a level, what's the difference? If their ML model uses a transformers-based architecture vs not, what difference does it make?
Traditional fraud-detection models have quantified type-i/ii error rates, and somebody typically chooses parameters such that those errors are within acceptable bounds. If somebody decided to use a transformers-based architecture in roughly the same setup as before then there would be no issue, but if somebody listened to some exec's hairbrained idea to "let the AI look for fraud" and just came up with a prompt/api wrapping a modern LLM then there would be huge issues.
