Good morning. It's Wednesday, August fifth, twenty twenty-six.

Story one: SpaceX just hit a milestone you might not expect from a company still trying to catch its own rocket mid-air. At three forty-two AM Pacific, a Falcon 9 lifted off from Cape Canaveral carrying three AST SpaceMobile BlueBird satellites — Block 2 this time, the ones with the massive phased-array antennas for direct-to-cell LTE from orbit. That launch was number fifty on Florida's Space Coast this year alone. All providers combined. Fifty orbital launches before August. The booster came back cleanly, by the way. The real headline though came out of the earnings call hours later: Musk confirmed Starship Flight fourteen will attempt the first operational orbit deployment this month, with a first-ever landing catch on the tower pad. Not an ocean splashdown — they're going for the chopsticks. That's a huge step up from Flight thirteen's soft Indian Ocean landing.

Story two: Anthropic is making headlines for something other than model releases today. Major outlets are running with the story of "Project Panama" — Anthropic's practice of buying physical books in bulk, prioritizing rare and out-of-print titles, then destructively scanning them by running the books through an automated cutting machine that shreds the spine. You get the digital text, the physical book becomes kindling. The Guardian ran an opinion piece calling it out. Snopes did a fact-check. Tom's Guide got a statement. Here's what makes it legally interesting: a federal judge already ruled the scanning itself is fair use, because you can transform lawfully purchased copies. But the settlement with authors and publishers that followed covered over four hundred thousand eligible works sourced from a pirated library — a separate issue from the scanning operation. The court allowed Anthropic to keep the digital files from its cutting process. So the physical books are gone, but the data stays. Book lovers are calling it book burning. Copyright lawyers are calling it precedent. Both might be right.

Story three: Qwen three point eight Max — Alibaba's new two-point-four trillion parameter model — is generating a lot of heat, and not just for the scale. The API launched yesterday at two dollars per million input tokens, positioning it as a direct competitor to GPT-5.6 Sol and Claude Opus 5. But the interesting part is what comes next: Alibaba confirmed the open weights are landing on Hugging Face next week. Combined with Kimi K3's open-weight release last week, Moonshot's aggressive pricing, and DeepSeek's V4 Flash at fourteen cents per million tokens, Chinese labs are now flooding the open-model space while the White House voluntary safety testing framework explicitly exempts Chinese open-weight models from its requirements. Hugging Face's own CEO told CNBC that China is dominating the open-model segment right now. The price war is here, and the regulatory asymmetry is making it worse.

Story four: A fresh study from Drexel University published in Nature Machine Intelligence adds a twist to the humanoid robot trust question. Researchers found that people quickly develop trust in expressive, socially engaging humanoid robots — but when those robots make mistakes, trust collapses much faster than with less expressive machines. The study used the ergoCub robot in a collaborative lifting task, measuring biomechanical stress and human partner adaptation. It's a reminder that as companies like BYD prepare to debut their Xiao Di humanoid in dealerships this month, and as the FCC's sweeping ban on foreign-made advanced robots reshapes what hardware can enter the US market, the social design of these machines matters as much as their technical capability. Expressiveness is a double-edged sword.

That's all for today. See you tomorrow.