Hacker News Reader

There are no visible watermarks, but model makers can use steganographic codes to identify outputs from their own models.

Text-to-Image Models Leave Identifiable Signatures: Implications for Leaderboard Security

This is true, however LMArena does employ some methods to mitigate attempts to manipulate the leaderboard, see https://openreview.net/forum?id=zf9zwCRKyP