Congress Is Suddenly Waking Up to the AI Doomsday Threat46m ago Don’t ask ChatGPT 'What should I do?' — this simple prompt helps you make better decisions yourself54m ago Extreme Networks’ Agent ONE Coworker moves AI networking from dashboards to answers56m ago Introducing ChatGPT for Financial Services1h ago Tesla unveils driverless Cybercab in Japan for 1st time2h ago The AI safety vibe shift2h ago Sam Altman met with top power utilities about securing the electrical grid. He offered one possible solution: OpenAI's cyber services.2h ago OpenAI Is Open to Slowing Cutting-Edge AI, CEO Sam Altman Tells Staff2h ago Tencent-Backed Chipmaker Enflame Jumps 188% in Shanghai Debut2h ago Salesforce introduces Enterprise AI Harness, AI Control Plane3h ago The Race for the Next Big Weight-Loss Drug Is On—and China Wants In3h ago How Social Media Platforms Are Identifying AI Content3h ago DeepSeek launches V4.1-Flash and retires V4-Pro, its flagship model3h ago OpenAI Wants to Know if an AI Industry Slowdown Would Even Be Legal3h ago

Congress Is Suddenly Waking Up to the AI Doomsday Threat46m ago Don’t ask ChatGPT 'What should I do?' — this simple prompt helps you make better decisions yourself54m ago Extreme Networks’ Agent ONE Coworker moves AI networking from dashboards to answers56m ago Introducing ChatGPT for Financial Services1h ago Tesla unveils driverless Cybercab in Japan for 1st time2h ago The AI safety vibe shift2h ago Sam Altman met with top power utilities about securing the electrical grid. He offered one possible solution: OpenAI's cyber services.2h ago OpenAI Is Open to Slowing Cutting-Edge AI, CEO Sam Altman Tells Staff2h ago Tencent-Backed Chipmaker Enflame Jumps 188% in Shanghai Debut2h ago Salesforce introduces Enterprise AI Harness, AI Control Plane3h ago The Race for the Next Big Weight-Loss Drug Is On—and China Wants In3h ago How Social Media Platforms Are Identifying AI Content3h ago DeepSeek launches V4.1-Flash and retires V4-Pro, its flagship model3h ago OpenAI Wants to Know if an AI Industry Slowdown Would Even Be Legal3h ago

AI Research Machine Learning7 min reading time

Safety for Whom? Refusing the Right Subset of a Topic, Not the Whole Topic

Hugging FacePublished 3 days ago

A new paper introduces a method for large language models to refuse only specific harmful subsets within a topic, rather than rejecting entire topics, enabling more precise safety controls tailored to deployment needs.

#llamaguard-3 #model-safety #or-bench #prompt-refusal #xstest

More in AI Research

AI Research4 min read

An Anthropic researcher just quit, saying OpenAI and Anthropic are 'gambling with our lives'

Covered by 13 sources2d ago

AI Research10 min read

Anthropic Details Disrupted Claude Misuse Across Seven Harm Areas

Covered by 4 sources9h ago

AI Research4 min read

OpenAI asks Congress for mandatory national AI safety rules before it adjourns

Covered by 2 sources20h ago