Good point. To add an exception to the original article then, perhaps the only general method to detect fraud is the trivial method: assume everything is fraud.

In practice though for most goals it's probably better to trade off Type 1 vs Type 2 error.