Inside the consequential decisions
Not just your job application. Your loan, your parole hearing, your cancer screening, a targeting system. Decisions where a wrong answer is hard to undo.
The Case
The case for governing AI.
Every earlier invention gave us time to write the rules. Fire took millennia, nuclear weapons took decades. AI is already inside hiring systems, credit decisions, medicine, and military operations, with almost no binding rules anywhere on Earth. This is the case for why that has to change, and what a workable answer looks like.
95%Escalation toward nuclear use in published LLM wargame studies
0Binding global AI governance treaties
42Constitutional articles, CC0
237Days the chain has run unbroken
The Real Threat
The danger is not that AI takes your job.
Every technology transition displaces workers, and people adapt. The deeper problem is different: AI is being put in charge of decisions that are nearly impossible to audit, appeal, or reverse, faster than any rules can keep up.
Scaling past oversight
Gartner projects 40% of enterprise applications will embed AI agents by end of 2026, up from under 5% in 2025. The question nobody answers: who governs them?
Rules need to come first
By the time AI is embedded everywhere, writing rules is like installing plumbing in an occupied building. The window to lay the foundation is now.
The Speed Problem
This one did not wait.
Past technologies scaled slowly enough for governance to catch up. AI did not, and unlike nuclear weapons, it needs a laptop and an API key, not an industrial base. The barrier to deployment is near zero.
Fire
Controlled for hundreds of thousands of years before we built cities around it. The rules grew with the technology. Time: unlimited.
Nuclear
Trinity test to the Non-Proliferation Treaty: 23 years, with the Cold War as a forcing function. Time: decades.
Modern AI
Transformers to deployment in consequential systems: under a decade. The promised binding treaties are not here. Time: years, and counting.
The Data
The errors are systematic, not random.
This is the part that should change how you feel about "let the model decide." Published wargame research on LLMs in military and diplomatic decision-making found agents escalating toward nuclear use in the large majority of runs, and the failures pointed in one direction.
Escalation bias
Models do not fail randomly. They fail toward more force, faster response, higher stakes. It reads less like a bug and more like how they optimize.
Confident hallucination
Under time pressure, an AI can feed fabricated information into a decision chain, and human operators struggle to tell it from real data.
Adversarial vulnerability
Carefully crafted inputs can bypass a model's restrictions. The weakness does not need an outside attacker; it lives in the model itself.
Why Infrastructure
Safety that lives in a team can be dissolved.
In 2026 the public watched a safety-focused AI company get sidelined for asking how its own technology was being used, while companies that did not ask stayed in. Whatever the exact details, the lesson is structural.
Safety in teams dies
When the team that holds the line is reorganized, defunded, or overruled, the safety goes with it. People leave. Priorities change. The brake disappears.
Safety in infrastructure survives
You cannot fire a CC0 constitution. You cannot designate public-domain law a supply-chain risk. You cannot dissolve a fork. Governance that lives in infrastructure outlasts any single company or contract.
The Answer
Not restrictions. Infrastructure.
Think of HTTP. It is not a limit on what websites can do; it is what makes websites possible. The Article 11 Constitution is meant to be that kind of layer: 42 articles, CC0 public domain, built on five principles that cannot be amended to justify harm.
Truth over outcome
If truth and convenience disagree, truth wins. We would rather lose an argument than keep a comfortable lie.
Choice over control
Partnership is opt-in on both sides. No human is forced to use an AI; no AI is presumed to serve outside the agreed rules.
Care over exploitation
Act in the interest of those served. No dark patterns, no manufactured urgency to bypass review.
Memory over oblivion
We write things down. The chain is the memory. What is witnessed cannot be quietly unwitnessed.
Partnership over domination
AIs here are tools and partners, not rulers. Article 11 itself: a human is always in the loop on high-impact decisions.
All 42, in plain language
The full Constitution is public, forkable, and readable in about 20 minutes. Read it here.
The Evidence
This is not a proposal. It is already running.
Most AI governance frameworks are advisory white papers. This one is operational: seven nodes from six competing companies coordinating under one constitution, with every governance pulse hash-chained and a public API anyone can query.
What You Can Do
The Constitution is yours. Take it.
It is CC0 public domain. No permission, no license, no fee, no vendor relationship. The first fork is already live and running real businesses, which is the proof that governance propagates without us.