Today in AI — 6 April 2026
Today's top AI news — curated links and commentary on the stories that matter for product builders.
The price tag for the AI buildout got more concrete today. It's counted in layoffs, in creative rights, and in the trust people place in systems they can't fully audit.
The bill comes due
Oracle began cutting up to 30,000 employees via mass 6 AM emails to redirect cash toward a $156 billion AI data centre push. Goldman Sachs projects semiconductor revenues will surge 49%, with the four major hyperscalers approaching $600 billion in combined AI capex this year. The spending is real. So is the human cost of funding it.
- Oracle cuts up to 30,000 jobs to fund $156 billion AI data center push — CNBC
- Goldman Sachs forecasts 49% semiconductor revenue surge driven by AI demand — ANI News
The ownership fight
The WGA locked in expanded AI protections covering training data licensing in a new four-year deal. GitHub, going the other direction, will train on Copilot users' code by default from 24 April — met with near-universal backlash. Meanwhile, Bloomberg reports 41% of code globally is now AI-generated, with 92% of US developers using AI tools daily. AI consumes human work to produce more AI-assisted work. Who captures the value is still being negotiated.
- Writers Guild reaches four-year deal with studios including expanded AI protections — Deadline
- GitHub will train on Copilot users' code by default starting April 24 — The Register
- Vibe coding is the AI trend fueling a new kind of developer FOMO — Bloomberg
Video generation reshuffles
Sora's 26 April shutdown is triggering a land grab across the $2.4 billion AI video market. Kling 3.0 undercuts on price, Google's Veo 3.1 Lite targets high-volume developers at half the standard cost, and Netflix open-sourced VOID — which erases objects from video and rewrites the surrounding physics, preferred over Runway 65% of the time.
- Kling, Runway, and Vidu are the video generators set to replace Sora — Bloomberg
- Google launches Veo 3.1 Lite, its cheapest video generation model yet — 9to5Google
- Netflix open-sources VOID, an AI model that erases video objects and rewrites the physics — The Decoder
Trust gaps
Ledger's CTO warns AI is driving crypto exploit costs toward zero after $1.4 billion in losses last year. The Washington Post — whose own AI podcasts invented quotes within 48 hours of launch — published an ethics framework for journalists using AI tools. And Anthropic confirmed Claude Mythos is in early access after a data leak forced the announcement. More capability arriving faster than the guardrails to match it.
- AI is making crypto hacks cheaper and easier, Ledger CTO warns — CoinDesk
- How journalists can use AI ethically — The Washington Post
- Anthropic's Claude Mythos enters early access as its most powerful model yet — Fortune
AI on the shop floor
Virtual try-on startups are reporting 10% conversion lifts and 36–40% return reductions for fashion retailers. One of the cleaner ROI stories in applied AI right now.
The thread running through all of it: AI systems are powerful enough that the fights over who benefits, who pays, and who gets to opt out have moved from theoretical to contractual.