Archive · Ars Technica

LLMs respond differently to harmful prompts when AI watermarking is used

Global Published 17 September 2026, 23:33 · archived 18 Sep 2026

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

ZAZTEK archives the headline, summary and publication date so the story stays findable. The full article remains the publisher's. Read it at Ars Technica →