If I had to guess, the companies that trained these models added the "snitch" hooks to trigger logging for later review of the users activity.
Or perhaps its just an artifact of models following the general, well published, sanitized, standard advice we see everywhere.