# Renting your AI: you have the use of it, not the ownership > A company often has only the use of its AI. It will not own the model, but it can own what surrounds it: its corpus, and the freedom to switch models. Article · By Élie Sader, Network Effect · Published on November 4, 2026 A business owner signs up for an artificial intelligence (AI) subscription, the teams use it every day, and the owner says, fairly enough, that the company has "adopted AI". Now ask the question that lawyers, heirs to Roman law, put to any form of ownership. It brings three rights together: the right to use a thing, the right to profit from it, the right to dispose of it. Take one away and the ownership is no longer complete; leave only the use, and what you hold looks like a lease. Applied to AI, the definition is harsh. The company has the use, on terms it did not negotiate; the profit is uncertain; the right to dispose does not exist. The point is not to give up on AI. It is to know what you rent and what you own, because the balance sheet does not tell the difference. > Three rights make ownership. AI leaves you one. ## What does it actually mean to own something? In the tradition that comes from Roman law, down to article 544 of the French Civil Code, unchanged since 1804, ownership is "the right to enjoy and dispose of things in the most absolute manner". Lawyers break it down into three rights, which must be held at the same time. *Usus*: using the thing. *Fructus*: collecting its fruits, putting it to profit. *Abusus*: disposing of it, meaning changing it, selling it, pledging it as collateral, or getting rid of it. Let us keep three plain words: use, profit, dispose. Lose one and ownership is split apart; keep only the use, and all you hold is a precarious right: a lease, a concession. The tenant of a shop has the use of it and profits from it through the business, but cannot sell it or knock it down, and the lease may not be renewed. The tenant knows this, and builds the business accordingly. With a tool, the line is harder to see. Accounting records what the company bought, not what it controls: a subscription and an asset you hold look alike on an expense line. If the test works for a building, it works all the more for a tool the whole company now depends on. Let us apply it to AI. ## What do I actually own of my AI? Take the three rights one at a time. **Use: yes.** Your teams reach powerful models, through an interface or a software connection, under a contract they did not negotiate and that the provider can change. A model is the program that understands and writes; it belongs to whoever trained it. **Profit: rarely, and uncertainly.** You do benefit from the tool, of course: time saved, better documents. But who owns what the AI produces for you is legal ground still being mapped. And what the provider learns from your use, your corrections, your documents, under the terms you accepted, goes to the provider. Every correction adds to a capital that is not yours. **Dispose: never.** You cannot change the model, sell it, pledge it to your bank, or carry it as an asset. If it disappears tomorrow, or starts behaving differently, you have nothing to stand on. This is not a criticism of AI. It is a statement about structure. And when the exercise is done honestly across all of a company's tools, the result is always the same: the list of things it fully owns is shorter than the owner thought. ## Does it matter, if everyone is a tenant? Everyone is in this position, and many live with it. As long as AI helps write a letter faster, it does not matter. It starts to matter the day you build an advantage on it: a service, a way of answering customers, a way of organising work that only holds if the tool holds. Building an advantage on an asset you only rent is building on sand, not because the technology is fragile, but because the ownership is incomplete. The day the terms change, the price, the usage policy, the regulation, or the provider itself, you have no right of disposal to protect you. An image helps. Think of the small farmer given the use of a tractor, while the land and the machine belong to others: he works faster, but the value of what he produces piles up with the owner. The business owner who "does AI" without asking what they own of it is playing the small farmer. The objection is serious, though, and deserves to be taken at its word. No small or mid-sized company will ever own a frontier model: training one takes means that only a handful of players in the world can gather. Demanding the right to dispose of the model is asking a craftsman to own his own foundry. Taken literally, the test produces nothing but paralysis: you adopt nothing, for fear of owning nothing. So the right target is not the model. It is everything around it. ## So what can I own? Move the target from the model to the ground it works on. What a company can hold outright, with all three rights, does exist. - **Its corpus**: its data and its knowledge, kept in readable files, in-house. The corpus is the structured body of what the company knows, which the AI comes to read in order to serve it. - **Its written processes**: the way work really gets done, which a competitor cannot copy by buying the same subscription. - **Its institutional knowledge**: what AI amplifies but does not replace. - **Its instructions**: the rules and guidance it has given its agents, recorded in-house rather than buried inside the tool. An AI agent is a program that acts on its own: it reads, prepares, carries out. These four things have one thing in common: they are pieces of the company's intangible capital, its knowledge of its customers, its craft, its processes, its weak signals, its dealings with the real world. It is the only thing in the whole AI picture on which all three rights can be fully held. The model is rented; intangible capital is owned, or lost. We describe its layers in [Intangible capital](https://networkeffect.fr/en/intangible-capital/). And there is a new kind of right of disposal: portability, the ability to change model or provider without losing anything. You do not dispose of the model; you dispose of your own freedom to replace it. Resilience does not come from absolute ownership, which is out of reach, but from that ability to leave. The test then becomes a diagnostic tool. For each critical asset in the company, draw three columns: use, profit, dispose. Tick what is real, cross out what is illusory. Three dependencies show up almost every time: - **know-how that lives in one person's head**: the company profits from it but does not own it; it belongs to whoever carries it, and will leave with them; - **a tool with no way to export** what you put into it: the right of disposal has been confiscated; - **a market held through a single relationship**: the use is precarious, a concession in all but name. ## What does Bitcoin teach about disposal? Bitcoin holders learned in one sentence what Roman law said in three words: not your keys, not your coins. A balance with a custodian is a right of use; the keys, the secret that alone can move the funds, are the right of disposal. Fifteen years of failed platforms have reminded anyone who forgot. That is the other face of intangible capital, digitised capital: it can only be owned through the keys, just as the corpus can only be owned if it is kept in-house. Same rule, same architecture of keys. The second lesson is about the cost of leaving. Mediocre systems do not last because people believe in them, but because getting out costs too much. The digital lord does not own the land; he owns the access, and his walls are terms of service nobody reads. The cage is gilded, and it closes quietly, as you pour your knowledge into it. Measuring your dependence on an AI therefore means measuring your exit cost: how much time, money and lost knowledge would it take to switch provider tomorrow morning? Choosing a tool looks like a reversible decision; it becomes irreversible as you put what you know into it. ## Where do I start? Three moves, in this order. **First, the corpus and the instructions in-house**, in files any model can read, today or ten years from now. It is the only part of AI on which you can hold all three rights, and it is the part that makes the difference between your company and the one next door. **Then, the model, taken for what it is**: a tool you can replace. Local when your hardware allows it, from a provider under contract otherwise, and never locked in with a single one. That is what we set up with [Infrastructure you own](https://networkeffect.fr/en/services/infrastructure/), with a written procedure so that you do not depend on us either. **Finally, the switch drill**: having actually changed model once, so that you know you can. An ability never exercised is a hypothesis. The point is not to own everything. It is to know what you rent, and to own what sets you apart. ## What you really own You will never own your AI model, and you do not need to. Own your corpus, your instructions, your processes and the ability to change model: that is full ownership in the age of AI, and it is within your reach. To equip yourself with tools that belong to you and keep the ability to leave: [Infrastructure you own](https://networkeffect.fr/en/services/infrastructure/). To know what to write down, and where to keep it: [Intangible capital](https://networkeffect.fr/en/intangible-capital/) and [CORPUS](https://networkeffect.fr/en/services/corpus/). ## Sources 1. French Civil Code, article 544 (text unchanged since 1804, apart from spelling): its successive versions on [criminocorpus.org](https://criminocorpus.org/fr/legislation/droit/versions/763/544/pdf/). ## Further reading - [Tools you own](https://networkeffect.fr/en/services/infrastructure/) - [The model will never be your advantage](https://networkeffect.fr/en/model-commodity/) - [All resources](https://networkeffect.fr/en/resources/)