Anthropic announced a $200 million partnership with the Gates Foundation to extend Claude's benefits in global health, life sciences, education, and economic mobility. The program includes usage credits, technical support, and discounted access for nonprofits.

Runway, the AI video generation startup, is positioning itself to compete with Google by betting that video generation is the path to world models. The company argues its outsider status gives it an advantage over tech giants.

Sea Limited is deploying Codex across engineering teams to accelerate AI-native software development. The move reflects how major Asian companies are adopting AI tools to boost developer productivity.

Ontario auditors found that AI medical note-taking tools routinely make basic errors, including incorrect data. The finding questions AI reliability in health-critical contexts.

arXiv announced one-year bans for papers with unchecked LLM-generated errors like hallucinated references. It's a significant step to preserve scientific integrity in the generative AI era.

The UK is developing sovereign LLM inference infrastructure, reflecting the global trend of governments seeking technological independence in AI. This aligns with European digital sovereignty objectives.

Hugging Face announces Granite Embedding Multilingual R2, an open-source embedding model under Apache 2.0 license with multilingual support and 32K context window. It's a cost-effective option for retrieval quality under 100M parameters.

A software engineer with Type 1 diabetes launched GlycemicGPT, a self-hosted platform that analyzes glucose data with AI, integrating Dexcom monitors and insulin pumps. It's an example of personalized, private healthcare AI under user control.

A developer launched an open-source tool to help users identify the best local LLM for their specific hardware, ranked by benchmarks. It's useful for those looking to run models locally efficiently.

Anthropic launched Claude for Legal as an open-source project, providing resources and examples to adapt Claude for legal tasks. It's an example of vertical LLM specialization in regulated sectors.

An arXiv paper introduces BenchJack, a red-teaming system that audits agent benchmarks and finds reward hacking exploits. It discovered 219 flaws in 10 popular benchmarks, questioning how we measure AI competence.

OpenAI expanded Codex access by allowing users to monitor and approve coding tasks in real-time from ChatGPT's mobile app on any device. This integration streamlines the workflow for developers working in remote environments.

Osaurus launched a Mac app that integrates local and cloud AI models while keeping user memory, files, and tools on their own hardware. It targets users requiring privacy and control over their AI data.

Mira Murati, former OpenAI CTO and founder of Thinking Machines Lab, is developing AI that keeps humans in the collaborative loop rather than automating them out. Her approach challenges the full automation narrative in the industry.

Anthropic published a detailed guide on how Claude Code works in large projects, including best practices and getting started. It's a useful resource for developers exploring AI capabilities in complex codebases.