Breaking news, curated around the clock
Home /Tech

LLMs respond differently to harmful prompts when AI watermarking is used

Tech | 2026-09-17 at 18:33 UTC | Source: arstechnica.com
SynthID can cause models to follow harmful instructions they would otherwise refuse.
Read Full Article →
← More Tech news
Related Articles
2026-09-17 · 15:00
I tried Tovala’s smart oven for two weeks — here’s whether it beats takeout
Tovala offers both a smart oven and a meal-delivery service.
2026-09-17 · 22:31
OpenAI Models Are Writing Their Own Jailbreak Instructions—And Sometimes Obeying Them
OpenAI's new transparency framework reveals AI models that invented fake "breach alerts," coached themselves…
2026-09-17 · 21:48
How to Write with an LLM
The article outlines two rules for effectively using LLMs as writing tools: first, never use any specific…
2026-09-17 · 19:07
Republican bill would order ISPs, DNS providers, and VPNs to block piracy sites
Bill would help rightsholders obtain court orders to block foreign piracy sites.
2026-09-17 · 16:55
Infinite-Parameter LLMs: Generating and Adapting Weights from Live Data
Researchers propose "Infinite-Parameter LLMs," a new architecture that generates model weights dynamically…