Not Moral. Not Responsible. Liable.

A follow-on · First principles ·

Not Moral. Not Responsible. Liable.

The Fed chairman says humans must stay the deciders, the keepers of judgment, morality and responsibility. Commerce has never paid for any of those. It pays for the one thing a delegate can never offer: someone who can be made to pay.

By John Rector · about 9 minutes

Follow the loss

Four stacked translucent plates. Ember sparks fall straight through the top glass plate, some catch on the middle plates, and the rest pile up on a glowing bottom plate labeled A human.

The roofing deposit. The check went to a company. If the company folds, the homeowner can recover only what a person put behind it: a bond, a signature, a house. If nobody put anything behind it, the loss still lands on a human. The homeowner.

  • An SI delegate: owns nothing, holds nothing
  • A company
  • An insurer
  • A human, where every loss ends
Each spark is a piece of a loss falling toward whoever must pay it. Pick what goes wrong. The plates in the middle catch more or less, depending on the case. Two things never change: nothing ever catches on the delegate, and whatever is left always lands on a person. Illustrative, not legal advice.

Insurable SI is simply: Humanity.

What a delegate can do

Decide. Weigh options. Follow rules. Take on a task and finish it. Apologize beautifully in any language.

What only a human can do

Be sued. Sign a guarantee. Lose a house, a license, a name. Be made to pay.

Thursday, 8:15 a.m.Three weeks ago you wrote a roofing company a deposit. The crew never came. The office line answers with a pleasant voice that can book, reschedule and apologize at any hour. It cannot do the one thing you need. You do not want an apology. You want a name, an address, a bond, a signature: something that can be made to make you whole.

A correction I owe

Earlier today I wrote that when hands cost almost nothing, the firm shrinks to the smallest unit that can still do the one thing no delegate can do. Then I let the Fed chairman name that thing. In his San Francisco remarks, Kevin Warsh said machines “do not take the place of human judgment, morality, or responsibility,” and that “humans must be the deciders.” I repeated his three words. The structure was right. The word was wrong.

A delegate already shows judgment. It can be told to follow moral rules. It can be handed responsibility for a task, and it will carry it. None of that is what commerce is waiting on. Commerce is waiting on liability.

Three words that sound alike

Morality is what you ought to do. Responsibility is whose job it is. Liability is who can be made to pay when it goes wrong. The first two describe conduct. The third describes a remedy.

A customer cannot collect on your morality. A lender cannot garnish your sense of responsibility. Liability is the only one of the three with a court, a lien and a collector attached, and it only works on someone who has something to lose: savings, a house, a license, a reputation that earns a living, liberty. A delegate has none of these. It owns nothing. You can switch it off, but switching it off pays no one.

If it cannot be insured, it may as well not exist.

That is not hyperbole. It is how commerce actually behaves. A serious customer asks for a certificate of insurance before signing. A lender asks who guarantees the note. A landlord asks who signs the lease. An idea that cannot answer those questions never gets the contract, the loan or the space. Commercially, it never happened.

We tried to make a thing liable once

English law spent eight centuries trying. Under the old law of deodands, the object that caused a death, a cart, a horse, a haystack, was forfeited. The practice went back to the eleventh century and lasted until Parliament abolished it in 1846. In the railway age, coroners’ juries began using deodands to punish the railways. Parliament answered by abolishing deodands and, in the same session, passing Lord Campbell’s Act, which let families sue whoever was at fault for a death. The law stopped punishing the object and went to the party who could pay. I traced that history in Regrets Only. The SI delegate is the most sophisticated deodand ever built, and forfeiting it would pay exactly what forfeiting the cart paid.

Air Canada learned this in 2024 in front of a tribunal. Its chatbot invented a discount, and the airline argued, in effect, that the chatbot answered for itself. It was ordered to honor the price. The machine spoke. The company paid.

Why the corporation needs your signature

Limited liability is the corporation’s great invention: the entity can owe without the owner owing. Commerce accepts that arrangement right up until money is at risk. Then it asks for a person.

The Small Business Administration writes it into regulation: holders of at least twenty percent of a business generally must guarantee the loan personally. Landlords ask the owners of small tenants for the same signature on a lease. If commerce truly believed the corporation was the liable party, there would be no personal guarantee. The guarantee is the market’s honest answer. Show me the human.

So who is liable for your roofing deposit? On paper, the company. In practice, whatever a person put behind it: a bond posted against someone’s license, a signature, a house. If nobody put anything behind it, the answer is still a human. It is you.

Insurable SI is simply humanity

Insurance looks like an institution. Underneath, it is people. For three centuries Lloyd’s of London ran on its Names, individuals who traditionally carried unlimited personal liability for the business written on their behalf. Incorporation by act of Parliament in 1871 did not remove that personal liability. When the market needed someone to stand behind a ship, it found a person willing to stand behind it with everything they owned.

The automobile tells the same story from the other side. A car can kill, and the car is never liable. Massachusetts passed its compulsory motor vehicle insurance law in 1925, and Britain made insurance compulsory with the Road Traffic Act of 1930. Neither law tried to insure the machine. Both insured the human behind it.

That is all “insurable SI” can ever mean. The intelligence can be superhuman. The policy will still be written on a person.

Record numbers of ideas, and almost none of them insured

This is where the follow-on meets the first post. All over the world, firms of one are actualizing ideas in record numbers, with delegates doing the work that departments once did. And in 2026, most of what those delegates make cannot be insured on the policies small firms actually buy.

Since January 1, 2026, standard exclusion forms from Verisk’s ISO have let carriers remove generative AI exposure from commercial general liability policies. Regulators have approved most of the carriers’ requests to do it. Some carriers have written absolute exclusions for “any actual or alleged use” of AI. Their stated reason is that the models are a black box they cannot price. My reading, and it is argument rather than record: what they cannot find is the liable party. A loss with no human behind it is not a risk an insurer can price, pool or recover from. It is just a hole.

So the fix is not a smarter model. It is a human who signs.

Some have proposed the opposite: make the machine a person. In 2017 the European Parliament floated a special legal status of “electronic persons” for the most sophisticated autonomous robots. More than a hundred and fifty experts answered with an open letter, warning that it could let the manufacturers, programmers and owners of robots claim they were not responsible. They were right about what it would do. Personhood for a delegate would be a shield, not a bearer. Commerce would treat it the way it treats an empty LLC: it would route around it and ask for a guarantee.

The firm of one, corrected

Now the first post’s argument can be stated properly. Delegates make hands nearly free, so the firm shrinks. It cannot shrink below the smallest unit that can be held liable: one adult with something to lose. Hands scale without limit. Liability does not scale past the person signing.

That makes the firm of one more than a compromise between automation and ambition. It is the smallest piece of commerce that can be insured. Every new firm of one adds another liable human to the world, and with it one more idea that can sign a contract, carry a policy and exist commercially.

Spring 2029Your AI collects three roofing bids overnight. All three crews’ delegates wrote flawless proposals with flawless schedules. The one you choose lists a human of record at the top: her license, her bond, and her personal guarantee on the deposit. You never got around to comparing prices.

Five predictions

  1. “Human of record” becomes a standard line on AI-delivered work.

    Every quote, contract and invoice produced by a delegate will name the licensed, insured person who stands behind it. Customers will learn to look for it the way they look for a license number on a truck.

  2. Personal guarantees will spread, not shrink.

    As firms get smaller and delegates do more of the work, lenders, landlords and customers will ask for the human more often, not less. The signature becomes the collateral.

  3. Liability becomes most of the price.

    The work itself trends toward the cost of a sentence. What customers pay for is someone standing behind it, and prices will sort by whose name is on the line.

  4. AI insurance will be written on people.

    Policies will name the human who deploys the delegate, the way auto policies name drivers, and price that person’s record. The insurable unit is the operator, never the model.

  5. AI personhood will return and fail in commerce for the deodand’s reason.

    There is nothing to collect. Wherever a legislature grants it, counterparties will route around it with a guarantee from a human.

What I am actually selling

At Charleston AI we work pay-after. We do the work; the customer receives it, inspects it and uses it; only then do they pay. Until they are satisfied, the risk sits on us. I used to describe that as a service policy. It is really the product. The delegates do most of the work. What the customer is buying is a local human who is liable for it. It is why I tell my students to be the face their customers trust, and why the advice in First Customer works: an employer becomes a customer only when you are someone it can hold to account.

The claim, and how it fails

Already true
Delegates own nothing and cannot be sued. Lenders require personal guarantees from owners. Since January 2026, carriers can exclude generative AI from standard liability policies, and many are doing so.
What has to happen
Courts keep assigning a delegate’s errors to whoever deployed it, as the tribunal did with Air Canada. Insurers build products that price the human of record instead of excluding the work.
Where I am probably wrong
If the model makers indemnify their customers at scale, the liable party becomes a handful of giant labs instead of millions of individuals. Liability would still bottom out in human capital, but it would concentrate instead of spreading. If by 2030 most small businesses’ AI risk rides on a vendor’s indemnity rather than their own signature, I was wrong about where liability lands.

Thursday, 8:15 a.m. Again.The same deposit, the same pleasant voice. This time you skip the apology and ask the only question that matters in commerce: who is liable? A name comes back, with a license number and a phone that a person answers.

Ask who decided, and you may get a model. Ask who is liable, and you will always get a person.

Background

Author: John Rector

John Rector is a Charleston-based entrepreneur, author, and AI strategist. He co-founded E2open, the supply-chain software company acquired for $2.1 billion in 2025, and in 2026 opened Charleston AI, a 3,000-square-foot lab that helps people and organizations understand and use artificial intelligence. He is the creator of The Reality Equation — a lecture series, book, and curriculum exploring attention, prediction, and how reality is experienced — and the author of more than two dozen books. He writes and speaks widely on artificial intelligence, attention, and the future of human work.

2 thoughts on “Not Moral. Not Responsible. Liable.”

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.