The White House Just Took Control of Who Gets New AI Models. China Just Gave One Away for Free.

HubSpot switched to Mintlify and saw 3x faster builds with 50% fewer eng resources. Beautiful, AI-native documentation that scales with your product — no custom infrastructure required.

Simplify Your Docs Today

Five stories today, and they all boil down to the same question: who actually controls AI, the companies building it, the governments regulating it, or whoever can build it cheapest? Today you got a partial answer on all three fronts. Let's get into it.

From elections and inflation to sports, tech, and more, Kalshi lets you trade on the real-world events you already follow. Buy “Yes” or “No” contracts based on what you think will happen, then earn returns if you’re right.

Pick a market, make your prediction, and put your knowledge to work.

Trade $25, Get Up to $500

_{Bonus credit varies from $15 to $500. Terms apply.}

The Trump administration set up a new system called Gold Eagle, run jointly by the Treasury, Homeland Security, and the Pentagon, that gets final say over which companies can access newly released frontier AI models, according to CNBC. It's already in effect. OpenAI's newest model, GPT-5.6 Sol, launched restricted to about 20 government-approved customers, and Anthropic's own most capable models, Fable and Mythos, were briefly suspended last month over export control concerns before being restored. Until now, OpenAI and Anthropic each ran their own internal vetting programs, called Daybreak and Project Glasswing, to decide who got access. Now the government sits on top of both.

Why it matters to you: If you build any part of your business around a specific AI model or tool, you're now one government review away from losing access to it, or watching a competitor get it first. This isn't hypothetical. It already happened to two of the biggest labs in the industry this month.

What to do about it: Don't build your business around a single AI vendor's latest model. Whatever workflow you depend on today, make sure it still works if that specific model gets pulled or delayed tomorrow.

Moonshot AI's new Kimi K3 model beat Anthropic's Claude Opus 4.8 on coding benchmarks while costing roughly 40% less to run, according to Moonshot's own numbers reported by the Associated Press, and the full model becomes downloadable for anyone by July 27. One startup founder told CNBC he moved all of his company's AI traffic to a cheaper Chinese model and watched his costs "crash to the ground."

Why it matters to you: This is the same story as yesterday's issue, still unfolding. Every week a free or nearly free alternative closes the gap with the AI tools you're paying for, the harder it gets for any vendor to justify a premium price tag. Stack that against the story above about the government gating US model releases, and open Chinese models start looking a lot more attractive to price-sensitive businesses, government access rules or not.

Anthropic is in early talks to lease up to $10 billion worth of Meta's computing power over two years, according to multiple reports. Either company can walk away before anything is signed, and both have declined to comment publicly. It's a smaller version of the $45 billion, three year deal Anthropic already struck with SpaceX for access to its Colossus supercomputers.

Why it matters to you: Two companies that compete head to head on AI products are apparently fine doing business behind the scenes when it comes to raw computing power. That tells you the AI compute shortage is real enough that even rivals will rent from each other rather than wait in line. If you've noticed AI tools getting slower or pricier during peak demand, this is part of why.

Google renamed NotebookLM, its tool for turning your own documents into summaries, podcasts, and Q&A, to "Gemini Notebook." Alongside the rename, it now connects directly to the Gemini app and Google Search, and each notebook gets its own secure cloud environment to run code and analyze data, not just summarize text.

Why it matters to you: If you've used NotebookLM to organize research, meeting notes, or client documents, nothing about what you built breaks, it just got more capable and a new name. If you haven't tried it, this is a good week to start. Feeding it your own documents and having it answer questions specifically about your business is one of the more genuinely useful free AI tools available right now.

Twenty nine countries, including Russia, Brazil, Cuba, Pakistan, and Indonesia, signed an agreement in Shanghai to create the World AI Cooperation Organization, a new international body headquartered in China aimed at setting global rules for AI. The US, UK, EU, and most of the wealthy Western economies did not sign.

Why it matters to you: The world is splitting into two camps setting the rules for how AI gets built and used, one centered on the US and its close allies, the other centered on China and offered as an open alternative to developing nations. Which camp your tools, vendors, and data end up in over the next few years won't be decided by you, but it will affect what you're allowed to use and where your data can legally go. Worth watching, not worth panicking over yet.

Here's today's throughline: control over AI is getting contested from every direction at once. The US government just inserted itself between AI labs and their customers. China is racing to make its models both cheaper and more available to the rest of the world, and building its own governance club to go with it. Rivals like Anthropic and Meta are quietly doing business with each other because nobody has enough computing power to go it alone. None of this changes what you should do on Monday morning. It does mean the assumption that any single AI tool you rely on today will look the same in six months is getting less safe by the week. Build flexibility into whatever you depend on. Stay skeptical. Stay curious.

If this saved you from digging through five different AI newsletters just to figure out what actually matters, forward it to one person who needs to stop guessing about AI too. Someone forward this to you? Subscribe so the next issue lands straight in your inbox.

Talk tomorrow,
Mark Shilensky

Follow me personally on Social Media:
Facebook: https://www.facebook.com/MarkShilenskyPage
Instagram: https://www.instagram.com/markashilensky/
Threads: https://www.threads.com/@markashilensky
LinkedIn: https://www.linkedin.com/in/shilensky/
X: https://x.com/markshilensky
TikTok: https://www.tiktok.com/@mark.shilensky
BlueSky: https://bsky.app/profile/markshilensky.bsky.social