Wall Street already loves ChatGPT. Now OpenAI is doubling down.1h ago Anthropic reveals rogue AI agents hate CAPTCHAs, just like you38m ago Cognition Adds Dioxus Team to Advance Devin Coding Agent39m ago Anthropic Details Disrupted Claude Misuse Across Seven Harm Areas44m ago OpenAI Releases GPT-6 Astra for Coding and Computer Use44m ago India’s Pocket FM doubles revenue run rate to $500M as AI powers 93% of audio content48m ago Andrew Tulloch, Poster Child for Out-of-Control AI Spending, Reportedly Quits Meta51m ago Build more natural voice experiences with GPT‑Live‑1 in the API1h ago Microsoft patches a record 974 flaws, and two are already under attack1h ago T. Rowe Price Expands Claude Across Investment Teams and Developers1h ago Val Bercovici, Chief AI Officer at WEKA – Interview Series1h ago Ben Jennings on the Bayeux tapestry – cartoon1h ago The latest Avatar: Seven Havens trailer gives us a deeper dive into the characters and story1h ago Agentic AI Is The New Attack Surface: How Can SAP Application Security Teams Repel It?1h ago

Wall Street already loves ChatGPT. Now OpenAI is doubling down.1h ago Anthropic reveals rogue AI agents hate CAPTCHAs, just like you38m ago Cognition Adds Dioxus Team to Advance Devin Coding Agent39m ago Anthropic Details Disrupted Claude Misuse Across Seven Harm Areas44m ago OpenAI Releases GPT-6 Astra for Coding and Computer Use44m ago India’s Pocket FM doubles revenue run rate to $500M as AI powers 93% of audio content48m ago Andrew Tulloch, Poster Child for Out-of-Control AI Spending, Reportedly Quits Meta51m ago Build more natural voice experiences with GPT‑Live‑1 in the API1h ago Microsoft patches a record 974 flaws, and two are already under attack1h ago T. Rowe Price Expands Claude Across Investment Teams and Developers1h ago Val Bercovici, Chief AI Officer at WEKA – Interview Series1h ago Ben Jennings on the Bayeux tapestry – cartoon1h ago The latest Avatar: Seven Havens trailer gives us a deeper dive into the characters and story1h ago Agentic AI Is The New Attack Surface: How Can SAP Application Security Teams Repel It?1h ago

AI Research35 min reading time

Claude Opus 5: Model Welfare

Don't Worry About the VasePublished Jul 27, 2026

Anthropic's Claude Opus 5 model achieved the highest scores on model welfare and alignment tests among recent models, though its performance may reflect strong test-taking abilities. The analysis highlights ongoing challenges and tradeoffs in improving AI model welfare as capabilities advance.

#ai-ethics #alignment #anthropic #claude-opus-5 #model-welfare

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

Google Pledges Over $15 Billion for AI Infrastructure in Finland

Covered by 3 sources1d ago