People in AI Towers Shouldn’t Throw Slop
Though I am not Catholic myself, I have been Catholic-adjacent my whole life. Some of my closest friends were raised in the Church (though most have lapsed). My mother taught for over a decade at a Catholic school. One friend’s father was the Australian ambassador to the Vatican, and another friend of mine is an acclaimed Vatican reporter who wrote a biography of Leo XIV not long after he acceded to the Chair of St. Peter. It’s safe to say Catholicism has always been in my orbit.
I launched a legal practice advising on AI governance in May and gave it a Latin name. In the same week I signed my second client, Pope Leo XIV published an encyclical on artificial intelligence with a Latin name of its own. Unsurprisingly, I had some thoughts.
I had recently named my firm Regulus, and it was in part because I’m a Leo and Regulus is the brightest star in the Leo constellation (and also in part because I’m 5′7″, and it means ‘short king’ in Latin). I did this before there was a Pope Leo with strong opinions about my industry. Now the brightest star in Leo and the man who took the name are aligning.
So recently, for the first time in my life, I sat down to read a papal encyclical. Two things landed strongly for me from Magnifica Humanitas, resonating with why I started this practice. At the moment His Holiness may be the loudest, sanest voice on AI out there.
Technology Is Never Neutral
The Pope writes that technology is never neutral, because it carries the characteristics of the people who build it, fund it, regulate it, and use it. That is less theology than a product fact. Anyone who has worked at a startup knows how much the product direction can be an extension of the personality of the founders and investors (for better or worse). Anyone who has read a model card, or terms of service, or the system prompt of any chatbot knows it too. Models are trained on a corpus of data that encompasses human thought with all its flaws and biases. The values are in there. At some point, a human chose them. The Pope just said the quiet part into a megaphone.
This is the entire premise of AI governance. The machine is not an oracle handed down from a neutral heaven and should not be treated that way. It is a stack of decisions made by humans with incentives. Part of my job is to help find those decisions, name them, and get them documented before they become a problem.
The Tool Is Not the Problem. The Operator Is.
Leo warns against the fantasy that a single language, even a digital one, can translate everything, flattening culture and meaning in the process. I say the same thing from a smaller chair.
Here is what I see often. The tool is not the problem. In spite of any and all good intentions, the operator is.
A layperson opens a chat window and types “make me a contract” and they get something that looks like a contract but contradicts itself internally. Not because the model is stupid, but more likely because the person cannot frame the question with enough nuance, cannot recognize the wrong answer, and has no position of their own with which to steer. They may still sign that contract and find themselves with no issues for years (until one day…)
A lawyer (or engineer, or sales rep, or Pope, or …) who understands the underlying issue, who comes to AI with a point of view, is more likely to get something useful out of the same tool without the same destruction of value. Because they are driving and they know what good looks like. They catch the confident nonsense before it lands in an operating agreement. The variable is judgment, and judgment does not come in the black box of a large language model.
Slop Has a Price
The contracts I get asked to fix are not AI failures. They are judgment failures. AI did not replace the lawyer. It removed the friction that used to stop a non-lawyer from generating ten pages of plausible, wrong, binding text in ninety seconds. It even slapped the label Privileged and Confidential in the header for good measure.
That label, by the way, is worth nothing. Earlier this year, in United States v. Heppner, No. 25 Cr. 503 (JSR) (S.D.N.Y. Feb. 17, 2026), Judge Rakoff held that a criminal defendant’s chats with an AI tool were not privileged and not protected work product, even though he used them to map out his own criminal defense. The tool is not your lawyer, and your inputs are not necessarily confidential once they live on someone else’s servers. Claude and ChatGPT do not decide what is privileged. Courts do. The government got to read all of it.
Heppner is the dramatic version of a real, measurable risk. For most businesses it shows up more quietly, and it lives in the documents that drive their work. After all, a contract is a system for allocating risk, and its value is in whether the parts hold together. That is exactly where unsupervised AI drafting fails.
The most common damage I see is internal contradiction. A payment term that fights the termination clause. A word that means one thing in section two and something else in section nine. An exhibit referenced but never attached. Worse is contradiction across the stack, where the NDA, the MSA, the order form, and the data agreement each say something different about liability, IP ownership, or governing law, with no order of precedence to break the tie. Worse still is lopsided risk allocation a non-lawyer cannot even see. The indemnity that runs the wrong way. The liability cap that protects the other side. The contractor IP assignment that never actually assigns anything, so you do not own the thing you paid to build. Each clause reads fine on its own. How it reads in context is the trap.
You probably won’t notice any of it at the time, but it will surface during diligence, financing, or a dispute, which is to say at the moment it is most inopportune and most expensive to fix. “It’s fine because the AI told me so” may not survive contact with a court, an investor, or a counterparty.
Diversity Is the Resource
There is another thread running through all of this, and Leo finds it in the story of the Tower of Babel. Babel did not only fall because it was godless. It fell because of uniformity. One language. Homogenization over communion. The fantasy that everyone could be flattened into a single output and nothing would be lost.
That is the AI slop problem stated in scripture. The pretense that one model, one voice, one answer fits every case. Diversity is the resource that breaks the spell. Diversity of training, of discipline, of judgment. Between law, technology, and operating, I happen to have a lot of it. It is the whole idea behind Regulus Law and Regulus Consulting.
Leo’s prescription is a firm relationship with God. I personally do not have one of those. Or maybe I do, if you let the word stretch to mean the universe, our collective humanity, or quantum physics. But when you boil it down, it means whatever you point to when you need something solid under your feet.
My approach to AI use in my own practice is, so far, more boring and more checkable. The position you take before you open the tool should be your terra firma. It is the thing the machine cannot give you, and the thing that decides whether the machine is useful or dangerous in your hands.
His Holiness ends by asking everyone to stop building towers and get their hands dirty on the construction site of our time. I will skip the part about the city where God and humanity dwell, but our instincts are the same. This era is being shaped right now, by the people selling you the tools, unless you insist on shaping it too.
I intend to help. So should you.
—
James Hoare is the founder and Principal Attorney of Regulus Law PLLC and a member of the New York State Bar Association’s Committee on AI and Emerging Technology.
