AI News Briefing — May 31, 2026

Anthropic announced a $65 billion Series H at a $965 billion valuation led by Altimeter, Dragoneer, Greenoaks, and Sequoia, surpassing OpenAI to become Silicon Valley's most valuable AI company. The round pushes its valuation near the $1 trillion mark.

A large-scale study spanning 208,000 participants and 26 million responses finds that the helpfulness training used to create chatbots degrades their ability to replicate human behavior, worsening with each model generation. Persona prompting with demographic profiles offered almost no improvement for individual predictions.

OpenAI's Codex app now runs on Windows 11 with 'Computer Use,' letting the AI autonomously control programs, test apps, and hunt bugs. Users can launch and monitor tasks remotely via the ChatGPT mobile app while away from the PC.

Genesis AI released Genesis World 1.0, a physics platform comprising a physics engine, the Nyx real-time path-traced renderer, the Quadrants Python-to-GPU compiler, and a simulation interface. It targets the model development cycle bottleneck by speeding up simulation-based evaluation of robotics foundation models.

Microsoft and Nvidia are reportedly partnering on AI PCs using Nvidia chips as the main processor, with Dell and Surface devices expected at Computex and Build. Microsoft is also building software, likely based on the OpenClaw framework, to run AI agents locally on Windows in a fresh attempt after Copilot+ PCs underperformed.

Salesforce claims migrating its entire dev org to Anthropic's Claude Code with no token limits yielded 79 percent more pull requests per developer and 5 percent fewer incidents in April 2026, including a 231-day migration cut to 13 days. The figures cannot be independently verified and fuel debate over whether agentic coding is a true revolution or a tech-debt liability.

Attackers are abusing the chat-sharing features of ChatGPT and Claude to distribute malware, disguising shared conversations as error messages or install guides. Because the content is hosted on trusted domains, it evades many security tools.

Nous Research's open-source Hermes Agent added a Tool Search feature to address MCP tool schemas bloating context windows, where deployments with 34 tools averaged 45,000 tokens per turn with about half consumed by schemas. Anthropic evals reportedly show accuracy rising from 49 to 74 percent on Opus 4.

Building on yesterday's Social buzz around Terence Tao's comments, Mathematician Terence Tao argues AI could introduce division of labor to mathematics for the first time, enabling large AI-supported teams ('industrial mathematics') rather than lone researchers mastering every step. He maintains humans remain essential for inspired conjectures.

A leaked Meta memo details an AI hardware strategy including an AI pendant, 'supersensing' glasses, and enterprise wearables, as the company seeks commercial payoff after heavy AI spending with limited returns. It frames hardware as Meta's next big bet following stalled open-source and research-to-product efforts.

Pope Leo XIV's first major papal teaching warns about AI threats to jobs, war, and the environment, with Anthropic co-founder Chris Olah seated beside him at the Vatican ceremony. Critics question whether the partnership amounts to genuine engagement or reputational 'Vatican-washing' by an AI firm.