The story of how AI entered its political era - and what it costs you.
AI has stopped being only a technology story. From here on, what your AI can do - and who it answers to - is decided as much by politics as by engineering. Someone else’s hand is now on the switch.
Political intervention was expected once AI reached a certain level. Aschenbrenner wrote about it in Situational Awareness, back in 2024 - years before Washington caught up.
“Somewhere around 26/27 or so, the mood in Washington will become somber. People will start to viscerally feel what is happening; they will be scared (…) Slowly at first, then all at once, it will become clear: this is happening (…) In one form or another, the national security state will get very heavily involved.”
Enterprises have been choked by the token appetite of agentic AI, which runs around the clock. Many have capped usage and are hunting for ways to keep the bill under control.
In a single year, Chinese models took the lead in token spend at up to 30× less the cost. The share of US closed models fell from 70% to 30%. China now holds 47% of the token market.
Fable came and went. An AI system cut off from the entire world in a single order - the first time export controls hit a model directly, not the hardware underneath it.
Raw model scores have converged. What separates the leaders now is the scaffolding built around them - and open weights have crossed the threshold where that’s enough. The Open Source Claude Moment has arrived.
“Open-weight models have crossed a threshold. GLM 5.2 beat a frontier agent on a bare prompt, at a sixth of the cost, with the option to run fully in your own environment. For a lot of security teams, that’s an attractive option.”
“Zenith took GPT-5.5 from 5th place to 1st, ahead of Claude Fable, by building the right harness around the model instead of swapping in a bigger one.”
Chinese labs allegedly harvest intelligence by firing millions of questions at US models and training their own on the answers. The method is called distillation.
Anyone can bury a hidden trigger inside a model’s parameters before release. It behaves normally, node by node, until it doesn’t. No standard safety check catches it in advance.
Washington’s worry may not be what we have today, but what comes next: recursive, self-improving models that build the next version of themselves.
Two battles will be fought simultaneously: keeping Chinese models constrained and keeping American intelligence contained.
Know Your Customer checks - the kind familiar from banking - may become standard for AI as models grow more capable. Access could soon mean showing ID.
A recent US Supreme Court ruling confirmed the president can fire the watchdogs meant to check him. Your EU data, hosted on US servers, can now be claimed no matter where those servers sit.
Europe might be forced by the US to pick sides. Say yes, and European hospitals, banks, and ministries answer to one office in Washington. Say no, and the fallback is open models nobody can fully audit. If you had to choose sides, which one would you pick?
US and EU legislation both arrive next month. From here on, your AI will be shaped by policy as much as by technology.
Europe needs enough of its own working compute, hosting and models so business can keep running if either side flicks the switch. That’s not sovereignty. That's a backup generator. But it gives it something to negotiate with.
Read the essay at nowable.tech →AI, power, and the architecture of what’s next. A few sharp reads a month, no noise.
See you in your inbox.
Something went wrong — try again or go to nowable.substack.com