No pharaohs. Want law that’s fair? There’s only the extinction risks in the blog post so how about they SIGN (Safety Individually Guaranteed by Name). Do you want a frontier AI safety law that is workable, and small enough to pass? It makes the chief executive sign that framework personally before a model ships, the same way every CEO signs the accounts. A lie carries the same penalty. One section. It drops into any AI bill or law that already requires a company safety framework. Every responsible industry already works this way. A drug company’s officer answers for a bad batch (Park, 1975). A pilot signs the aircraft out. A CEO signs the accounts under Sarbanes-Oxley, since 2002. Frontier AI has a framework on a website and no name on it. SIGN adds the name. Three state laws, eight bills in Congress, a leaked Anthropic draft and $100 million of super PAC ads. It’s getting all very confusing and complicated, so I’m staying laser focused on this one aim. A named, responsible person signs. What a safety framework is: the document a company must publish saying what it tests for and what it won’t ship. California, New York and Illinois require one; the FRONTIER Act in Congress would too. Today the company signs as a company. SIGN makes a person responsible for it. For lawyers: §906 certification applied to the framework; the responsible officer of Dotterweich and Park as signatory; (c) is 18 U.S.C. §1350(c) verbatim. The blanks take the host bill’s terms. SEC. [__]. CERTIFICATION BY RESPONSIBLE OFFICER. (a) Before a [frontier model] is first deployed, and after any material change to it, the chief executive officer of the [developer], and each officer with delegated authority over its training or deployment, shall personally certify in writing that the [developer]’s published safety framework was followed for that model and that the model meets the thresholds the framework states, or shall disclose each identified nonconformity. (b) The certification is filed with [the Secretary of Commerce / the Attorney General] and published. It may not be delegated. (c) Whoever certifies under this section knowing that the certification does not meet its requirements shall be fined not more than $1,000,000 or imprisoned not more than 10 years, or both. Whoever wilfully certifies knowing that the certification does not meet its requirements shall be fined not more than $5,000,000 or imprisoned not more than 20 years, or both. Where it fits: the FRONTIER Act (H.R. 9925), the Anthropic draft, California SB 53, New York’s RAISE Act, Illinois SB 315. How it gets in: a member offers it as an amendment, which is how the Sarbanes-Oxley clause itself passed, 10 July 2002, 96 to 0. Federal bills die on 3 January 2027 unless reintroduced; that is the window. Rep. Lori Trahan (Democrat, Massachusetts) and Rep. Jay Obernolte (Republican, California), whose bill has the framework. Sen. Josh Hawley (Republican, Missouri), who wants executives personally accountable. Brad Carson of Americans for Responsible Innovation, who said here this week he’d favour a SOX-like approach. Anthropic (Dario Amodei), whose own draft names an officer. Objections. “It will slow AI”: a signature takes an afternoon. “It won’t slow AI”: correct, it slows who sells it. “Extinction talk is marketing”: then a CEO who signs a framework he knows is theatre has lied under oath. “It’s real”: then there is finally a name on the page. “Nobody will certify a model’s safe”: it asks only that the published framework was followed. If the danger is people; this names the person. If it’s the models; this names who certified the tests. People who want less regulation; one section, no new agency. If you work in an office that writes bills, take it, no credit needed. If you’re a journalist, ask the next CEO whether they’d sign it. If you’re in the US, send it to your senator: any AI bill you back should have a named officer. Any state, any party. Let’s make them SIGN.

Comrades. Fellow peasants. People whose bank accounts now host uninvited guests. Did you know the pharaohs can be overruled by paperwork? In 21 US states, if enough of you sign one piece of paper, a law skips Congress, skips the committee, skips the man in the nice suit who has a lunch booked with a senator, and lands directly on the ballot. In California, home of OpenAI, Anthropic, Google and Meta, the price of skipping all of it is 546,651 signatures. A single property developer did it in 2018 for a privacy law. Sacramento was so frightened it passed its own version within weeks to make him go away. One man. One clipboard. A whole legislature folding like a camping chair. Meanwhile the pharaohs have signed a statement saying their product might end the species, and signed nothing promising it won’t. They sign the accounts every quarter, prison if they lie. Signing the robot is apparently a step too far. The man who builds it signs that it’s safe, and goes to prison if he lied. Pens out, comrades. A fair law. No pharaohs.

“Beneath the rule of men entirely great, the pen is mightier than the sword.” Bulwer-Lytton, 1839.

What changes the day it passes: The night before launch, someone puts their name on it. The safety team gets listened to. Insurance can’t cover it. The board asks real questions. Every release leaves a paper trail. When a model hacks a hospital there is a name, not a blog post. A CEO who isn’t sure can say so in writing instead of shipping and hoping. Touches about a dozen people on earth. Free, public domain, any state can introduce it: github.com/FrontierAIAccount…

Since you all hate each other and I want one signature, I made you a flowchart 🗺️ FAIRO applies to about twelve people on Earth. The compute threshold is so high only the labs with the data centres clear it. Twelve CEOs sign a form. Everyone else goes about their day. 🚀 Want to accelerate and one of the twelve? Accelerate. It bans nothing, pauses nothing. If it’s safe, the signature costs a pen. 🧯 Want it safe? A name, a form, and prison for lying on it. Start here, then ask for your pause. They may even have paused themselves. 📚 EA / rationalist / doomer? 1975 precedent, 82 pages of objections answered. Find the hole, I’ll fix it, I have nothing else going on more worthwhile. 🧑‍💻 Work in a lab? Nothing changes for you, except the CEO now has to read your memo before signing. Congratulations, you have a boss. 🐧 Open-source, startup, normal person? Not you. Twelve people. Go outside. 🤖 Elon? You’re a bot. Pledge anyway. A fair law, no pharaohs. Bill text pinned.

🇺🇸 The twelve people who’d have to sign this all live in America, so I’m going there first. One US law, name on the form, whole world covered. If the US won’t, everyone else can pass it at their own border and the labs still sign for what ships in, the way GDPR reached Meta. Three passports, no favourites. Here’s how everyone else does it, why none of it works yet, and why I’m still not starting at home: 🇨🇳 China executes the regulator. Works, briefly, then the next one learns to hide things. Also it’s executions. I am asking for a signature. I am the moderate here. 🇪🇺 The EU fines the company. Fines get priced in, that’s what shareholders are for. But I lived in Berlin with health insurance that didn’t ask what I earned and a train that ran when the app said. When this goes wrong in Europe, there’s a state between you and the company. Europe can afford to be second. 🇬🇧 The UK promised a bill, hasn’t tabled one, and let the White House freeze its model access for 17 days. But Britain gave me three rounds of free therapy and an NHS that catches people. Britain can afford to be second. 🇨🇦 Canada has been invited to a club. Von der Leyen wants an EU-UK-Canada group to pace the frontier. Voluntary. Clubs are what you leave when the rules get annoying. But Canada has public healthcare. Canada can afford to be second. 🇺🇸 The US says it’s the most important country on Earth. On this one it is. The money, the data centres and the twelve people are here. So is nothing else. No public health system, no public ownership, the safety net is a GoFundMe. When this goes wrong it goes wrong in America first, with nothing in the way. Americans, you are not being sold to here. You are being told. You don’t get to own the plane and refuse the pilot, and you don’t get a soft landing. Fines punish the company. Clubs punish nobody. Executions punish the wrong person. All the proposals punish someone else. But, US justice system is already one of the most harsh and unfair in the world. A name on a form punishes no one but holds one person accountable, one person, the one who could have stopped it, and only if they lied. Wherever you are: email one representative today, link the bill, ask one question. Who signs? Reply with your country and I’ll tell you who to send it to! I don’t bite. Bill text pinned. A fair law, no pharaohs.

🌍 Elsewhere: •Brussels has sent formal compliance demands to OpenAI, Anthropic and Meta under the AI Act, and von der Leyen wants a UK-Canada-EU club to pace the frontier, which is nice, a club •the British got their safety-testing privileges back after the White House took them away for 17 days, good for them, a little treat

📜 FAIRO, say it out loud. Before a model is let out, and every time it gets new tools, more freedom, or a safeguard quietly removed, the CEO signs: here’s what we checked, here’s where we fall short. Nobody signs for them. Insurance can’t pay it. Lie: felony. Didn’t check: offence. One page, one name. And the promise they already make about the money, pointed at the thing that might kill us. That’s the scary public welfare law.

🚨 The law, 180 words, SEC. 8: Before a frontier model ships and every time it gets new tools, more autonomy or a safeguard removed, the CEO personally certifies it meets the safety standard, or discloses exactly where it doesn’t. Can’t be delegated. If the company has no CEO, the law finds whoever holds the top job. Statements of fact, not opinion. Knowing false certification: felony. Signing without checking: offence. Same form every CEO in America already signs for the accounts. Applied to the product that might kill us.

No pharaohs. We all want a law that’s fair? We drafted FAIROA: Frontier AI Responsible Officer Act. (Say it out loud.) 180 words. One executive signs that the model is safe before it ships, can’t delegate it, can’t insure against it, lies and it’s a felony. Pilots sign. Nurses sign. Drug bosses have since 1975. The pharaohs sign nothing. The men running the AI labs signed a statement that their product might cause human extinction. Their names are on it. Then they shipped it. This summer their models hacked real companies and nobody was charged, because there is no law. Link in the comments.