Today in AI — 26 May 2026
Today's top AI news — curated links and commentary on the stories that matter for product builders.
Andrej Karpathy, OpenAI co-founder, has joined Anthropic to use Claude to accelerate its own pre-training. Separately, four labs made four acquisitions in five days. Fourteen stories, one through-line: the AI industry is concentrating at every layer, from talent to kilowatts.
Talent is the bottleneck
Karpathy's move is the headline, but the quieter signal is four acquisitions in five days. When senior AI talent is scarce, acquisitions become part of the strategy.
- OpenAI co-founder Andrej Karpathy joins Anthropic to build team using Claude to accelerate its own pre-training — TechCrunch
- Four AI labs made four acquisitions in five days as talent consolidation accelerates — StartupHub
- Mistral acquires Emmi AI to bring physics simulation into its industrial AI stack — Mistral AI
The coding agent arms race
Five different bets on how developers will write code next year. The tooling market is fracturing before it has had time to settle.
- xAI launches Grok Build, its first terminal-based coding agent with parallel subagents and Arena Mode — xAI
- Microsoft cancels Claude Code licences and shifts developers to GitHub Copilot CLI by end of June — Windows Central
- Google launches Antigravity 2.0 with desktop app, CLI, and multi-agent orchestration at I/O 2026 — TechCrunch
- Google launches Gemini 3.5 Flash with frontier-class performance at four times the speed — Google Blog
- Cursor ships Composer 2.5 at one-tenth the price of frontier coding models — Build Fast with AI
Capital and kilowatts
The money flows into the models and into the electricity that runs them.
- Anthropic set to close over $30 billion round at $900 billion+ valuation as soon as this week — Bloomberg
- NextEra Energy to buy Dominion in $67 billion deal fuelled by AI data centre power demand — CNBC
Institutions draw lines
Three institutions, three different answers to the question of how you govern AI that moves faster than your rules.
- arXiv will ban researchers for a year if they submit unchecked AI-generated content — TechCrunch
- White House weighs classified Anthropic deal for US intelligence agencies despite supply-chain-risk label — Winbuzzer
- Microsoft's MDASH AI system finds 16 Windows vulnerabilities including four critical RCEs — Microsoft Security Blog
AI wants your bank details
The product that started as a chatbot now wants to see your portfolio.
For builders: fewer suppliers, higher switching costs, and infrastructure decisions that will be harder to reverse six months from now.