LLMs respond differently to harmful prompts when AI watermarking is used

(arstechnica.com) by joozio | Sep 18, 2026 | 0 comments on HN