Hedy AI announced Local AI Processing on May 13, 2026, enabling meeting analysis to run entirely on user devices without sending conversation data to remote servers.
American chip export controls have forced Chinese AI researchers to extract 4-7x more intelligence per compute unit, narrowing the performance gap despite a three-year hardware handicap.
Researchers developed a consensus-based framework to compare medical AI segmentation models using NLST chest CT scans, revealing variable performance across anatomical structures.
Princeton-affiliated researchers published findings in Nature showing government-coordinated media content appears 41 times more frequently in AI training data than Wikipedia, systematically biasing model responses toward state narratives.
A Nature study reveals state media control leaves detectable traces in AI model behavior through training data, with Chinese-language prompts producing 75% more favorable government responses than English.
The University of Utah will deploy new AI computing systems and modernize the Utah Population Database through a $33 million state investment aimed at accelerating medical research.
Fields medalist Timothy Gowers reports GPT-5.5 Pro produced doctoral-level number theory research in under two hours, while Anthropic introduces a memory-refinement process called "dreaming" for Claude agents.
Norwegian and Greek researchers release an open-source Wi-Fi cyber range prototype using software emulation and LLM-powered scenario generation for 802.11 security training.
The world's first AI art museum will debut a multi-sensory exhibition combining machine learning with 12 unique scent profiles diffused through smart devices.
Group-IB introduces Prevyn AI as the cognitive core of its Unified Risk Platform, using 11 specialized agents to transform threat intelligence research from hours to minutes.