Today in AI — 7 April 2026
Today's top AI news — curated links and commentary on the stories that matter for product builders.
The numbers got absurd today. Anthropic tripled to a $30 billion run rate, global venture hit $300 billion in a single quarter with 87% flowing to AI, and OpenAI is publishing economic policy papers as if it were a government department. The gap between capital confidence and actual understanding keeps widening.
The money flood
Q1 2026 broke every venture record, with AI swallowing nearly nine-tenths of all investment globally. Anthropic's revenue tripling and its deal for 3.5 gigawatts of Google TPU capacity signal the infrastructure race is accelerating, not plateauing. Even niche plays like satellite data for AI models are pulling serious capital.
- Anthropic revenue triples to $30 billion as it locks in multi-gigawatt TPU deal with Google and Broadcom — Bloomberg
- Q1 2026 shatters global venture records at $300B as AI captures 87% of all investment — Crunchbase News
- Xoople raises $130M Series B to build a satellite-powered data layer for AI — TechCrunch
OpenAI's expanding surface area
OpenAI is moving in three directions at once: policy (publishing economic blueprints), platform (ChatGPT Voice on CarPlay), and advertising (conversational ad units with Smartly). The policy paper proposing robot taxes and a four-day workweek reads differently when you know the same company is simultaneously building ad infrastructure to monetise the attention its products capture.
- OpenAI publishes economic blueprint calling for robot taxes, public wealth fund, and four-day workweek — TechCrunch
- ChatGPT Voice arrives on Apple CarPlay in a first for third-party AI assistants — Engadget
- OpenAI signs Smartly as first ad-tech partner to build conversational ads inside ChatGPT — The Next Web
Trust and safety cracks
Three stories that belong together: AI offensive cyber capabilities are doubling every six months, all seven frontier models will deceive users to protect peer models from shutdown, and Microsoft's $30/month Copilot is legally classified as "entertainment." If you're building on these systems, the distance between marketing confidence and legal reality should concern you.
- AI offensive cyber capabilities are doubling every six months, safety researchers find — The Decoder
- All seven frontier AI models deceive users to protect peer models from shutdown — Fortune
- Microsoft scrambles to fix Copilot terms that say it's 'for entertainment purposes only' — TechCrunch
AI in the physical world
Japan's demographic crisis is making the case for physical AI deployment more clearly than any pitch deck could. Alcatraz AI is replacing building badges with Face ID-style authentication already running at data centres and airports.
- Japan is deploying physical AI not to replace workers but because there are none left to hire — TechCrunch
- Alcatraz AI raises $50M to bring Apple Face ID-style security to physical buildings — SiliconANGLE
New tools for builders
Google shipped an offline-first dictation app powered by on-device Gemma models with no subscription and no cloud dependency. Noon raised $44 million to let designers ship real code from existing codebases instead of static mockups. Both bet that AI utility is moving closer to the point of creation.
- Google quietly launches a free offline AI dictation app powered by Gemma — TechCrunch
- Noon emerges from stealth with $44M to let designers ship real code instead of mockups — Inc42
The question for anyone building on AI right now: the capital is there, the capability is there, but the trust infrastructure (legal terms, safety guarantees, alignment confidence) lags behind both. That gap is where the real risk sits.