Today in AI — 4 May 2026

Today's throughline: the physical and legal world is catching up to AI's pace. Whether it's memory chips, subway ad campaigns, or Academy Award rules, institutions are scrambling to draw lines around what AI systems can claim, consume, and create.

Taking without asking

• Microsoft caught sneaking 'Co-Authored-by Copilot' into VS Code commits — even with AI turned off — The Decoder
• OpenAI turns on marketing cookies by default for free ChatGPT users — The Decoder
• 'This is Fine' creator says AI startup Artisan stole his art for subway ads — TechCrunch

Three stories, same structure: an AI company helped itself to attribution, artwork, or browsing data, then waited to see who'd notice. The opt-out-by-default playbook is getting crowded.

Capability versus permission

• In Harvard study, AI offered more accurate emergency room diagnoses than two human doctors — TechCrunch
• Oscars ban AI-generated actors and AI-written scripts from awards eligibility — Variety

AI keeps proving competence while institutions decide where competence isn't the point. The question isn't whether AI can do the work; it's who gets to decide when that matters.

The physical squeeze

• Apple kills the $599 Mac Mini as AI data center demand drives a 90% DRAM price surge — Fortune
• Coatue launches Next Frontier to buy land for AI data centers, targeting tens of billions — TechCrunch

AI's appetite for hardware is reshaping consumer electronics. The supply chain tax on everyone else is becoming visible at the checkout.

Agents in the wild

• Cloudflare and Stripe launch a protocol that lets AI agents create accounts, buy domains, and deploy code — Cloudflare Blog
• Meta acquires Assured Robot Intelligence to build humanoid robots — TechCrunch

Software agents are getting wallets; hardware agents are getting bodies.

Power and position

• Replit CEO reveals $1 billion run rate, Apple App Store fight, and says he won't sell — TechCrunch
• Pentagon tech chief says Anthropic is still blacklisted, but Mythos is a 'separate national security moment' — CNBC
• 21 European AI startups to watch beyond Lovable and Mistral — TechCrunch

Replit revealed a billion-dollar run rate (up from $2.8 million two years ago) and is ready to fight Apple in court. The Pentagon confirmed Anthropic remains blacklisted even as the NSA uses its Mythos model, calling it a "separate national security moment."

If you're building products today, budget for the world asserting its claims — on your supply chain, your defaults, and your training data.

---