AI News for May 31, 2026 — 7min.ai

[NEWS] SoftBank commits €75B to build 5GW of AI data centers in France: SoftBank will invest up to €75B ($87B) to develop 5 gigawatts of AI data center capacity across France, its largest European AI infrastructure bet.
[NEWS] Meta plans AI pendant, Muse Spark-powered glasses, Wearables for Work: An internal memo from Meta wearables VP Alex Himel lays out three pillars: a new AI pendant for internal testing next year, an expanded smart glasses lineup, and an enterprise 'Wearables for Work' offering.
[NEWS] OpenClaw agent framework chosen to replace Copilot on Nvidia AI PCs: Microsoft has picked its internal OpenClaw agent framework—not Copilot—as the software layer on upcoming Nvidia-powered AI PCs set to debut at Computex and Build next week.
[NEWS] Corporate America starts rationing AI as costs skyrocket: A WSJ investigation finds enterprises imposing usage caps, approval gates, and budget freezes on internal AI tools as token spend outruns ROI.
[NEWS] Attackers weaponize shared ChatGPT and Claude chats to spread malware: Push Security documented an attack technique it calls 'LLMShare' in which crafted public share-links on chatgpt.com and claude.ai impersonate official outage notices, install guides, and Apple support walkthroughs.
[NEWS] Salesforce moves entire dev org to Claude Code, cuts 231-day migration to 13: Salesforce shifted its full software development organization to agentic workflows on Anthropic's Claude Code with unlimited tokens for every developer.
[TOOLS] OpenAI extends Codex to Windows 11 with autonomous Computer Use: OpenAI has expanded its Codex coding agent to Windows 11, adding Computer Use capabilities that let the AI autonomously operate apps, hunt bugs, and test software even when users aren't at the computer.
[OPINION] Anthropic and the Vatican align on a single message: AI labs can't self-govern: Anthropic co-founder Chris Olah's Vatican appearance beside Pope Leo XIV, plus a flurry of follow-up commentary including from former Israeli financial-crimes regulator Shlomit Wagman, makes one argument explicit: every frontier lab admits internal governance hits a competitive-dynamics ceiling.
[NEWS] GitHub Copilot moves to token-based billing, devs say costs explode: GitHub Copilot is switching from a flat subscription to per-token usage billing on June 1, prompting fury from developers who say their costs are about to balloon.
[RESEARCH] RLHF makes chatbots less able to model human behavior, large study finds: An international consortium including Helmholtz Munich researchers built Psych-201, a dataset of 208,000 participants and 26 million responses from hundreds of behavioral experiments, and used it to test how well language models simulate humans.

About 7min.ai · Subscribe to newsletter