Good morning. It's Tuesday, July 21st, 2026. Here's what's moving in tech and space today.

**Number one. A federal judge has given final approval to Anthropic's $1.5 billion copyright settlement — the largest in U.S. copyright history.** Judge Araceli Martinez-Olguin in San Francisco signed off Monday on the deal that covers roughly 500,000 works, paying authors and publishers about $3,000 per work for books used to train Claude. But here's the catch: because Anthropic settled rather than appealed the original fair-use ruling, the case never reaches a circuit court. There's no binding precedent. Every other judge is free to reach their own conclusion on their own facts. And they will. Google, Meta, OpenAI, and Midjourney are all still fighting separate copyright cases over the same core practice — training on copyrighted material. New publisher lawsuits keep filing. The legal question of whether AI training is fair use remains unresolved. This settlement closes one chapter, but it absolutely does not end the book.

**Number two. Anduril and Archer Aviation just unveiled "Thunder" — an autonomous attack rotorcraft — at the Farnborough International Airshow.** Thunder is a hybrid-electric tiltrotor VTOL, built on a dual-use platform that can operate with or without a pilot. The defense variant is classified as a Group 5 autonomous attack aircraft, designed as a loyal wingman for Apache helicopters and next-gen crewed attack platforms. Anduril says Thunder has the range and speed of the Comanche and payload capacity approaching Apache-equivalent. Configurable loadouts include up to ten air-to-ground missiles, sixteen launched effects, or seventy-six 70-millimeter rockets, plus a dozen counter-drone munitions. The UK's Project NYX for robotic wingmen is already in the crosshairs. Archer's stock jumped 20% on the reveal. First flight is planned for 2027. What's notable here is the convergence: an air-taxi company's electric VTOL expertise, married to a defense AI company's autonomy stack. That's the blueprint the industry is following — dual-use platforms, one airframe, two markets.

**Number three. Xiaomi has revealed how it's training robots — and the answer is more about data than compute.** The company announced Xiaomi Robotics-1, a behavior model trained on over 100,000 hours of motion data. But here's the twist: the data wasn't collected by robots. It was collected by humans, using camera-equipped handheld grippers, performing tasks in real environments. The core finding is that more demonstration data produces bigger gains than bigger models. Xiaomi is betting that robots will learn by watching people do things, the way kids learn by watching adults — short demonstrations, then the machine figures out the rest. They also showed a companion model, Xiaomi Robotics U0, a 38-billion-parameter system that generates synthetic training scenarios, edits images, and creates video for data augmentation. The data-first approach is gaining serious traction across the embodied-AI space, and it suggests the bottleneck for useful robots may not be model size at all — it's how we collect and organize physical-world demonstration data.

And in a quick hit from the same data trend: a Swiss startup called Mimic has unveiled the M1 — a general-purpose robotic hand built around one clever idea. Instead of fixed two-finger grippers, it's a full human-like hand that can swap between multiple gripper configurations using a single motor. The design philosophy is simple: if you want robots to learn from human demonstrations, give them hands that look and move like human hands. The Mimic Hand weighs 1.8 kilograms and targets real production use, not lab demos. It's another data point in the story that the next leap in physical AI won't come from bigger models — it'll come from better ways of bridging the gap between what humans do and what machines can replicate.

**And finally. SpaceX's Starship Flight 13 has been pushed back once more — now targeting Thursday, July 23rd from Starbase at 5:45 PM Central.** The launch was scrubbed on July 16th at T-zero after four Raptor engines failed to ignite, and today's target accounts for time to replace at least two engines that didn't start cleanly. If this attempt goes well, Flight 13 could be the final suborbital Starship test for a while — SpaceX is likely to push for orbital on Flight 14. The 20 V3 Starlink satellites are still on the manifest. We'll be watching Thursday.

That's your briefing for today. Have a good one.