To identify a financial fraud, e.g. money laundry, AI deployment is not new. Given a large dataset, many machine learning techniques can identify patterns to help to identify these rare events.

Is it possible for some crime lords with bottomless resources to deploy an AI to outsmart the Anti-Fraud-AI (in short: AFAI)? In fact, this is quite easy to answer. The simplest solution would be to flood the financial transaction database with fake positives. An Anti-Anti-Fraud AI (in short: AAFAI) would first identify some fraud patterns, which are most likely identified by the AFAI, then do millions of fake transactions with similar patterns. Other solutions may exist as well.

So what should (or would) an AAAFAI do? and of course, an AAAAFAI do? can this continue infinitely?