LLMs respond differently to harmful prompts when AI watermarking is used

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

More in Policy & Safety

Policy & Safety

A US-China AI Hotline Won't Be Ready For a While Feed: Artificial Intelligence Latest · 8h ago

Policy & Safety

The Pope’s AI Guy Is Worried About ‘Cartel’ Behavior Among Big Labs Feed: Artificial Intelligence Latest · 11h ago

Policy & Safety

OpenAI extends cyber access to Ukraine for civilian defense OpenAI News · 16h ago

Policy & Safety

Sam Altman’s remarks at the United Nations Security Council OpenAI News · 17h ago

Policy & Safety

Microsoft disrupts AI-assisted platform that compromised 12,000 accounts AI - Ars Technica · 1d ago

Policy & Safety

Lawsuit demands OpenAI pay for new school after ChatGPT used in shooting AI - Ars Technica · 1d ago