How so?

Spam, for decades, has been a matter of just shoveling truckloads of emails out the door and hoping that one or two get a gullible match.

Blocking spam, for decades, has been a matter of heuristic pattern-matching.

I don't see how that is the same as "fighting LLMs with LLMs", or how it could be said to be the same as how spam is made and used.

It's analogous in the sense that tons of machine-submitted emails exist for the sole purpose of being mechanically triaged. The technology might be slightly different (it may not be LLMs through and through), but the pattern is the same.