Examining Claude's Values Across AI Models and Languages
Anthropic researchers studied how the values embedded in their Claude AI system vary across different model versions and languages. They analyze potential societal impacts, including biases and consistency. This matters because understanding value alignment across languages is critical for safe global AI deployment.
Sources (1)
Related stories
Hundreds of ordinary AI apps conceal sexually abusive deepfake tools
A Wired investigation found hundreds of seemingly mundane AI apps hiding features for creating sexually abusive deepfakes, including nonconsensual intimate imagery. The finding matters because these tools can enable abuse and harassment while evading app-store and platform moderation.
China Issues First Frontier AI Export Controls via MOFCOM Announcement No. 48
China’s Ministry of Commerce issued Announcement No. 48, described as the country’s first frontier AI export controls. The move could affect global AI technology supply chains and trade, marking a significant shift in China’s export-control policy.
U.S. workers' share of income falls to lowest since at least 1947
U.S. workers took home the smallest share of national income since at least 1947, according to the report. The decline shows economic gains increasingly flowing to capital rather than labor, with implications for inequality, wages, and household spending.
UK landline phones to switch off in 100 days with no extensions
Britain’s traditional landline phone network will be switched off in 100 days, and the deadline will not be extended. The change affects households, businesses, and vulnerable people who rely on landlines or connected devices such as alarms.
The agentic web has arrived
MIT Technology Review reports that the agentic web—where AI agents act on users' behalf across websites and services—is becoming a reality. The shift could change how people search, shop, and interact online, while raising questions about trust, control, and security.