Today in AI — 9 April 2026

Meta spent two years as AI's loudest open-source champion, then shipped its most competitive model as fully proprietary. The broader thread today: openness was the strategy for gaining ground, not for holding it. Meanwhile, healthcare is becoming the first sector where AI-versus-AI conflict plays out at real economic scale.

Meta's proprietary turn

Muse Spark launched as Meta's first closed model from its Superintelligence Labs, ranking fourth globally but leading every health AI benchmark. Wall Street rewarded the reversal with a 9% rally; the developer community that built on Llama's openness got an API waitlist.

• Meta debuts Muse Spark, its first proprietary AI model, in a ground-up overhaul of its AI strategy — TechCrunch
• Meta's new model is as open as Zuckerberg's private school — The Register
• Muse Spark ranks fourth globally but leads every model on health AI benchmarks — Artificial Analysis
• Meta stock rockets 9% on Muse Spark launch in biggest rally since January — Invezz

Healthcare's AI reckoning

Eli Lilly fired up pharma's most powerful AI supercomputer to compress drug timelines, while insurers and hospitals quietly admit AI scribes are inflating billing nationwide. One side creating value, the other extracting it.

• Everyone agrees AI scribes are increasing healthcare costs — no one agrees what to do about it — STAT News
• Eli Lilly inaugurates pharma's most powerful AI supercomputer with 1,016 Nvidia GPUs — Fierce Biotech

China's robotics capital surge

D-Robotics closed $150 million and Spirit AI pulled in $420 million backed by Lei Jun and Jack Ma. The speed and scale suggest Beijing-aligned capital sees humanoid robotics as the next front where industrial policy meets AI capability.

• D-Robotics raises $150M as Chinese embodied AI funding accelerates — The AI Insider
• Spirit AI raises $420M for humanoid robot intelligence backed by Lei Jun and Jack Ma — The AI Insider

Safety and security infrastructure

OpenAI published a framework for preventing AI-generated child exploitation material. Separately, Trent AI launched with $13 million to build security tooling for autonomous agents. Both point to the same gap: capable AI systems need dedicated safety engineering, not afterthoughts.

• OpenAI publishes a child safety blueprint to combat AI-enabled exploitation — TechCrunch
• Trent AI emerges from stealth with $13M to secure autonomous AI agents — SecurityWeek

Enterprise tooling bets

Narwhal Labs raised £20 million for fully autonomous enterprise communications; Patlytics raised $40 million for AI-powered patent analysis. Both betting AI can own entire business functions end-to-end.

• Narwhal Labs raises £20M and launches DeepBlue OS for fully autonomous enterprise communications — EU-Startups
• Patlytics raises $40M Series B for AI-powered patent and IP analysis — BusinessWire

If you're building on open-source AI models, today is the day to stress-test that dependency.

---