CommentShield: The AI Assassin of AI Spam

The author built a "poisoned watermark" prompt-injection that forces AI comment generators to output glyph gibberish and corrupts custom memory, claiming it reduced AI spam on their blog from about five a day to zero; the linked Medium article explains the technique, placement nuances, and how it exploits model prompt-handling to neutralize low-effort automated commenters.