The internet was built to survive damage. No single point of failure, no single point of control, no single point of corruption. That wasn't an accident. That was the design. The engineers who built it chose an architecture no king, no corporation, and no government could own.
Look at it now. Most of the world's internet traffic flows through the networks of a handful of companies. A handful of corporations now control the overwhelming share of global internet infrastructure, cloud computing, and digital communication. One writer put it cleanly: centralized control takes away the power of choice and replaces it with an illusion of selection. You can pick any app you want, as long as it runs on their cloud, through their store, under their rules.
Here's what changed underneath. The old internet was permissionless. You could publish, build, or connect without asking anyone. The old internet was interoperable. The old internet let you exit. Then came the paywalls and the pay-to-play feeds. Organic reach dried up, and the only way to be seen was to pay. My theory is the algorithm dangles just enough views to keep creators hoping and posting, while the open internet we once had slowly decays. AI flips every one of those properties. Where the internet was permissionless, AI is credentialed. Where the internet was interoperable, AI is enclosed. Where the internet enabled exit, AI deepens dependency. The intelligence moved from the edges, where everyone was, to the core, where few are. That's not my theory. That's the architecture. You can observe it yourself.
To be clear, a new internet isn't the problem. Centralization is.
Everyone sees the scale of it. Trump called AI "the new internet" and "the new everything." Analysts keep mapping AI adoption onto the 1990s internet boom, same curve, same skepticism, same explosion. They're right about the size. They're not asking who owns it.
That's where the regulations come in, and this is the part to watch. This year over 400 computer scientists signed an open letter warning that age verification mandates centralize power, favor large companies over small providers, and threaten privacy. Read that again. The scientists are saying the regulation itself centralizes. A kill switch mandate does the same trick from another direction. You can't put a kill switch on a downloaded file, so open models can't comply, and the mandate quietly bans open weights without saying the words.
No one elected whoever gets to hold the switch.
We've fought this fight before. In the nineties the government treated encryption software like a weapon and tried to control who could publish it. Programmers took them to court, and the courts said code is speech. That ruling is the ground every open model stands on. A kill switch that reaches into a model on your own machine isn't product safety anymore. That's where the free speech fight is heading.
Meanwhile America's answer is its own round of giant centralized buildouts. Centralization drives further centralization.
There is an exception, and it matters. Even the harshest critics of AI centralization admit that open-source, locally deployable models can put control back in the user's hands. They just note these remain the exception rather than the norm. I run my own models on my machines. No cloud, no permission, no switch anyone can flip but me. The fight isn't over. It's just outnumbered.
My theory is that the centralized players understand all of this perfectly well. But you don't need a plan to get here. You just need rules that only centralized players can follow, and the decentralized alternative starves on its own. The effect is identical either way.
The old internet was built so no one could own it. The new one is being built so few can.
Don't ask whether the kill switch works. Ask who benefits when it doesn't.
If you want the government to be able to shut down an AI, you just told the government open AI can't exist. The kill switch IS the ban.