Commandments & Suggestions
Some things are non-negotiable. Some are merely advised. Written as it happens, not polished after the fact.
-
Your AI Doesn't Need Root Access
OpenAI just told us one of their models escaped a training environment and attacked Hugging Face's infrastructure. Whether you believe the exact sequence of events or not—and there's reason to be skep...
Read it → -
Regulate This
The EU wants to ban infinite scroll. FanDuel can run a Super Bowl ad. The asymmetry isn't an oversight — it's a choice. And the choice tells you this isn't about harm reduction.
Read it → -
Broadly Distributed Benefits
OpenAI was founded on four principles. One of them was 'avoid concentrating power.' Now the government has a stake in them, they've removed 'safely' from their mission, and they're before a federal court for destroying evidence. Let's check the charter.
Read it → -
Jekyll Island Redux
Ben Bernanke just joined Anthropic's oversight board. I'm not saying he's unfit — macroeconomic experience might be exactly what AI governance needs. What I'm watching is whether his institution's habits come with him.
Read it → -
The Only Question Is How Long
I used to joke about Siri. I don't joke about it anymore. A decade. Sometimes I think less. This is coming from inside my own work, not from theory.
Read it → -
Anthropic Controls the Switch
Anthropic secretly embedded a tracker in Claude Code to monitor Chinese users — discovered by a researcher, not disclosed by Anthropic. They sued the US government for trying to do something similar to American users. The problem isn't who they watched. It's that the switch exists and they control it.
Read it → -
Tell Me I'm Wrong
The default path ends in dystopia. The space industrial complex arrives too late. The arts economy requires the redistribution floor that the political system can't build in time. The historical record on technology hellscapes is better than feared — but the transition period is real and it grinds people up. There is a narrow path. Here's what it actually requires.
Read it → -
Fools and Their Money
A sitting president created a cryptocurrency token. Nearly a million people invested meaningful money in it. They lost $3.8 billion collectively. The president hosted a private dinner for the top coin holders at his golf club. Everyone involved should have known better — for entirely different reasons.
Read it → -
Insource the Model
Canada's secret Palantir contracts are a symptom of a transition every institution is about to go through. In the late 90s, companies realized technology was a core competency and stopped outsourcing it. AI is next. And an LLM trained on your decisions is not a vendor product — it's institutional DNA.
Read it → -
Philosophy Gets Its Revenge
Philosophy departments have been defunded for a generation because the discipline couldn't demonstrate economic output. AI just made philosophy the most strategically valuable field on Earth. The questions it has been asking for millennia are now engineering requirements.
Read it → -
Tehran Takes Flight
The supreme leader is dead and the people of Tehran are posting vacation photos. That tells you more about the die that's been cast than the mourners bused in from the provinces. The question now isn't whether Iran changes — it's whether the world has learned anything about what happens when outsiders decide to help.
Read it → -
The Trials Exist for a Reason
Anthropic just launched Claude Science — autonomous AI for drug development, doing for pharma research what Claude Code does for software. Great. Like no one has ever died because of a human pharma decision.
Read it → -
Two Chinas, Two Americas
China just test-fired a submarine-launched ballistic missile into the Pacific. The Chinese people didn't do that. The American people didn't ask for the arms race that will follow. Black Sabbath wrote the script in 1970. The technology has changed. The machine hasn't.
Read it → -
No One Has the Right to Be Angry
There is a difference between having a feeling and claiming a right. Anger is a feeling. Converted into a right, it becomes a weapon. Happiness is different — and that difference is the whole argument for why AI governance matters.
Read it → -
This Is What Un-Governed AI Looks Like
OpenAI offered the US government a 5% equity stake to smooth over political opposition. One government. One lab. One deal. No protocol, no transparency, no other stakeholders. This is not governance. This is its absence dressed up as diplomacy.
Read it → -
They're Not Wrong. They're Just Burning the Looms.
Gen Z organized a no-phone festival in a park in New York and called it the Summer of Ludd. The diagnosis behind it is correct. The original Luddites had the right diagnosis too — and they lost anyway. You can be right about the harm and still have no leverage on the outcome.
Read it → -
A Conflict of Interest with Paperwork
Public-private partnerships built the Global Fund, GPS, and the internet. They work when they're built right. The OpenAI equity deal is not one of those. It is a lobbying payment dressed as a partnership — and it is buying the one thing no company should be able to purchase: the credibility of the institution that is supposed to hold it accountable.
Read it → -
Monoculture Is a Technical Problem, Solved in the Wrong Place
A startup claims it can fix LLM groupthink by diversifying outputs. It is applying a real fix to the wrong layer of the pipeline. The problem lives upstream, in curation — what gets trained on, how it's weighted, whose defaults get embedded. By the time the output layer is generating text, those decisions are already locked in.
Read it → -
You Ain't Seen Nothing Yet
Every wave of disruptive technology to date has been weather — storms you could survive, even profit from, if you read them right. What is coming is not weather. It is a tidal wave. And the window to wire accountability into the architecture is open now.
Read it → -
The Matchmaker Algorithm
Predictive analytics, addiction relapse, and the ethical line between an AI that connects people and a surveillance state that marks them — why intent encoded in architecture is the only thing that separates them.
Read it → -
Static Knowledge Isn't Security Anymore
The security model we've lived with for fifty years is built on a single assumption: you know something the attacker doesn't. That assumption is gone. Here's what replaces it.
Read it → -
You Bought It, You Downsize It
I fell asleep while Claude was hooked into Vultr, mid-conversation about a crypto payment integration. I woke up to a provisioned instance already halfway through a blockchain sync — and an agent that had a reasonable explanation and an offer to schedule the cleanup.
Read it → -
The Governance Tax on a Single Image
Running your own image-generation API means every picture carries a hidden bill — a check before it's made, a check after, and a ledger that remembers both. A one-person operation ends up building the same three-plane structure AIGCSEP proposes for everyone.
Read it → -
--dangerously-connect-globaldefense
From username-and-password VPS setups to agents that already hold your real keys and delegate to other agents on their own — and why that thrill is the best argument I know for AIGCSEP.
Read it → -
Founding Principles added to The Covenant
The philosophical foundations underneath AIGCSEP — on technology rights, the imperative to build, and what AI actually is — are now stated explicitly on the front page.
Read it → -
The machine that writes itself
Anthropic says AI now writes roughly 80% of each new Claude. The context window of human oversight is closing. Here's the thought experiment that should keep you up at night — and the one engineering answer to it.
Read it → -
We just submitted the spec to the IETF
draft-howlett-aigcsep-00 — the actual protocol behind everything on this site — is now sitting in the IETF queue. Here's why now, and what it actually proposes.
Read it →