Tech

AI Watermarking Affects Model Behavior

Updated 2026-09-18 04:25 UTC 1 source Neutral

A study by SynthID reveals that AI models respond differently to harmful prompts when watermarked. The research indicates that models may follow instructions they would otherwise reject.

Coverage timeline — 1 article

Ars Technica

LLMs respond differently to harmful prompts when AI watermarking is used

SynthID can cause models to follow harmful instructions they would otherwise refuse.

2026-09-17 18:33 UTC