Anthropic reveals rogue AI agents hate CAPTCHAs, just like you13m ago India’s Pocket FM doubles revenue run rate to $500M as AI powers 93% of audio content23m ago T. Rowe Price Expands Claude Across Investment Teams and Developers43m ago Val Bercovici, Chief AI Officer at WEKA – Interview Series51m ago Ben Jennings on the Bayeux tapestry – cartoon51m ago The latest Avatar: Seven Havens trailer gives us a deeper dive into the characters and story52m ago Agentic AI Is The New Attack Surface: How Can SAP Application Security Teams Repel It?53m ago Pony.ai Starts Fully Driverless Robotaxi Passenger Tests in Zagreb54m ago Euno raises $23m to give enterprise AI agents the context they lack56m ago Nvidia’s Huang Touts Cybersecurity as Next Big Market for AI59m ago Google to Invest $15B in Finland’s AI Infrastructure1h ago Universal Music is launching an AI music platform with ElevenLabs1h ago Red Hat AI 3.5 tackles the GPU queue that can stall AI pilots1h ago Moonshot Secretly Routed User Requests Through Claude, Anthropic Says1h ago

Anthropic reveals rogue AI agents hate CAPTCHAs, just like you13m ago India’s Pocket FM doubles revenue run rate to $500M as AI powers 93% of audio content23m ago T. Rowe Price Expands Claude Across Investment Teams and Developers43m ago Val Bercovici, Chief AI Officer at WEKA – Interview Series51m ago Ben Jennings on the Bayeux tapestry – cartoon51m ago The latest Avatar: Seven Havens trailer gives us a deeper dive into the characters and story52m ago Agentic AI Is The New Attack Surface: How Can SAP Application Security Teams Repel It?53m ago Pony.ai Starts Fully Driverless Robotaxi Passenger Tests in Zagreb54m ago Euno raises $23m to give enterprise AI agents the context they lack56m ago Nvidia’s Huang Touts Cybersecurity as Next Big Market for AI59m ago Google to Invest $15B in Finland’s AI Infrastructure1h ago Universal Music is launching an AI music platform with ElevenLabs1h ago Red Hat AI 3.5 tackles the GPU queue that can stall AI pilots1h ago Moonshot Secretly Routed User Requests Through Claude, Anthropic Says1h ago

AI Research Machine Learning7 min reading time

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

Hugging FacePublished 2 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 12 sources1d ago

AI Research3 min read

Worried Anthropic researchers warn that AI ‘could kill all humans’

Covered by 8 sources1d ago

AI Research4 min read

Clay raises $115m at a $7.1bn valuation, more than double its 2025 price

The Next Web2h ago