Hey from Tobias,

This was the week three governments reached into the AI market at once, and it should change how you think about the tools your workflows run on.

The US pulled Anthropic's two most powerful models overnight. China shipped an open model that nearly matches them at a fraction of the cost. Argentina started writing a law to let AI run companies with no humans involved. Here's my read, plus two free things I'm giving away.

A Government Just Pulled the Most Powerful AI Model

On June 12, the US government issued an export-control order, and Anthropic shut off Fable 5 and Mythos 5 for every customer worldwide that same night. Officials believed someone had found a way to jailbreak Fable 5 and unlock the cyber abilities of Mythos, its more powerful sibling. One directive, and hundreds of millions of people woke up without access.

The next day, China answered. Z.ai released GLM 5.2, an open model you can download and run yourself. It matches Opus 4.8 on most coding and reasoning tests and costs about 4.7 times less per token. Add the rumors that Anthropic may soon require ID checks to use its models, plus Europe's push toward verified identity for more of the internet, and the lesson is plain. If one order can switch off the model your business depends on, you don't own your stack. That's why I'm moving myself and several clients toward private, open-source AI we control.

Argentina Wants AI to Own Companies

Argentina's government submitted a bill on May 29 to rewrite its 1972 companies law and create the AI company: a business that can own assets, sign contracts, and pay taxes with no human owner, director, or shareholder required. President Milei has already met with Peter Thiel about it. Yuval Noah Harari and Microsoft's AI chief Mustafa Suleyman have both warned it's a mistake.

It sounds wild until you remember the limited-liability company was just as radical when Dutch merchants invented it in 1602 to share the risk of long sea voyages. That one idea unlocked four centuries of growth. Giving an AI its own legal shell is the same bet, placed again. Brilliant or reckless, it tells you where this is going: software that doesn't just do the work, it owns the business.

We Don't Need Bigger Models Anymore

The quiet story under all this noise is that raw model size has stopped being the point. GLM 5.2 just showed a cheap, open model can sit a hair behind the frontier. The real gains now come from how you connect models together: agents that plan, check their own work, and make the next run better.

I'm convinced we already have enough horsepower to reach AI that improves itself, without waiting for the next giant model. That's good news for you. The edge goes to whoever builds the smartest system around today's models, not whoever can afford the biggest one.

A Free AI Audit Skill I Built

I built an audit skill you can drop into Claude, or whatever AI assistant you already use. Install it once, point it at a website, and it does the digging for you. It's the same one I run on client work.

Then you just ask it something like this:

Swap in your own domain and your own question. It comes back with specifics, not vibes. Want it? Just reply to this email and I'll send you the SKILL file.

Own your stack before someone else decides you can't.

Tobias