◼ Archived Technology 🔮 holds ✗ LLMs respond differently to harmful prompts when AI watermarking is used 5 sources 6 articles v 3 4d ago