Daily AI Brief - Friday, May 15, 2026

Generated: 2026-05-15 Items: 69 new stories


Daily AI Brief — May 15, 2026

TOP STORY: Ontario Auditors Find Doctors' AI Note Takers Routinely Blow Basic Facts A provincial audit reveals that AI-powered medical transcription tools are consistently producing inaccurate clinical notes, raising serious patient safety concerns. The Register


Industry

PwC Deploys Claude Across Enterprise Functions - Anthropic expands its partnership with PwC to build AI-powered technology and reinvent core enterprise services for clients. Anthropic

Anthropic Partners with Gates Foundation for $200M - The two organizations join forces to direct AI capabilities toward global health and development challenges. Anthropic

Access to Frontier AI May Soon Be Gated by Economics and Security - A new essay argues that financial and geopolitical pressures will increasingly restrict who can access cutting-edge AI models. Anton Leicht


Tools

Codex Comes to ChatGPT Mobile - OpenAI brings its Codex coding agent to the ChatGPT iOS and Android apps, enabling on-the-go AI-assisted development. OpenAI

VS Code Agents Window Supports Local Models — With Caveats - The new VS Code "Agents window" allows local AI model use but still requires an internet connection and an active GitHub Copilot subscription. VS Code Docs

How Claude Code Navigates Large Codebases - Anthropic publishes a practical guide to Claude Code's architecture and best practices for using it effectively on complex, large-scale projects. Anthropic


Research

RewardHarness: Self-Evolving Agentic Post-Training - A new framework enables AI agents to iteratively improve their own reward models through post-training feedback loops. Hugging Face

STALE: Can LLM Agents Know When Their Memories Are No Longer Valid? - Researchers probe whether language model agents can detect and handle stale or outdated information stored in their memory systems. Hugging Face

WildClawBench: A Benchmark for Real-World, Long-Horizon Agent Evaluation - A new benchmark tests AI agents on realistic, extended tasks that better reflect the complexity of real-world deployment. Hugging Face

Unlocking Complex Visual Generation via Closed-Loop Verified Reasoning - A new approach uses closed-loop verification to improve the accuracy and coherence of AI-generated visual content. Hugging Face


Community

everything-claude-code - A community-built agent harness for Claude Code featuring skills, instincts, memory, and performance optimization. GitHub

Ollama adds Kimi-K2.5, GLM-5, MiniMax & more - The popular local model runner expands its supported model library with several new frontier and open-weight releases. GitHub

Dify hits 141K stars as production-ready agentic platform - The open-source platform for building and deploying agentic AI workflows continues rapid community growth. GitHub