Who Owns the AI Your Development Company Builds?
IP ownership in AI development contracts is more complicated than in traditional software — and the defaults often favor the vendor, not you.
You hire a software agency to build a website. You pay for it, so you own it. That is the normal deal. But the question of who owns AI you build is harder to answer. A custom AI project creates many separate parts. There is the trained model. There is the training data. There are the fine-tuned weights. There is the inference setup. And there is the application code. Each part is its own asset. Each one can be owned by a different party. It all depends on your contract. Most clients learn this too late. They find out after the project is signed.
This guide covers the main ownership questions in AI contracts. It shows you the risky defaults to watch for. And it shows you what to ask for before you sign.
What assets does an AI project actually create?
An AI project usually creates several separate deliverables. Each one raises its own ownership question. First is the application code. This means the backend API, the frontend, and the integrations. Next are the model weights. These are the trained parameters that hold what the AI "knows." Then comes the training pipeline. These are the scripts and tools used to train or retrain the model. You also get the training data, or the pipeline that processes it. Then there is the evaluation framework and the test sets. Last is the inference setup. Your contract should name each one clearly.
Model weights are the core IP asset. You cannot retrain, change, or copy the AI without them. If you do not own the weights, you stay tied to the vendor.
Training pipelines give you the freedom to retrain. Without them, you depend on the vendor for every future update.
Application code is where most "work-for-hire" clauses fit cleanly. It is often the only asset a basic contract names.
Owning the application is like owning a car with no engine. The model weights are the engine. Make sure your contract gives you both parts.
Common contract traps buyers miss
One risky default shows up often in AI contracts. The clause gives the code to you. But it keeps the model weights and training tools for the vendor. The reason given is usually "to operate and improve their services." This is common with vendors who use their own platforms. It is also common with vendors who reuse your data to improve models they sell to others. Here are more traps to watch for:
License-not-own clauses: you get a license to use the model, not the weights. The vendor can change the terms, raise the price, or end the license.
Training data reuse: clauses that let the vendor use your private data to train other models or improve their own.
Hosted-only delivery: you reach the model only through the vendor's API. You get no weights and no way to run it on your own.
No retraining rights: you own the weights, but the contract bars you from retraining or changing the model without the vendor.
What you should insist on
Your contract should give you a few clear things. You should own every model weight and parameter made during the project. You should own all of the application code. You should have the right to retrain and change the model on your own. The contract should bar the vendor from using your training data for anything outside your project. And you should get the model in a portable format. You should be able to run it on your own setup, or on standard cloud setup, without the vendor. A vendor may not agree to these terms. If so, find out exactly what you are buying and why.
The foundation model question
Many AI systems are built on top of commercial foundation models. Examples include GPT-4, Claude, Gemini, and Llama. The base model belongs to its creator, under its own license. You cannot own that layer. But you should own everything built on top of it. That means the fine-tuning you add to the base model. It means the retrieval system and the prompt work. It means the application layer. And it means your custom testing setup. Ask the vendor to split these layers clearly in the contract. To see how this shapes build choices, read custom AI vs off-the-shelf AI tools.
IP ownership ties closely to two other issues. One is vendor lock-in. The other is long-term operating cost. To surface these before you sign, see what to ask before hiring an AI development team and AI development red flags.
What if the vendor uses proprietary tooling to build my system?
Some vendors build on their own platforms. Others use internal tools that cannot transfer to you. This is not always a dealbreaker. But the vendor must disclose it. And you must price it into your choice. You are taking on vendor dependency. In return, you may get faster delivery or a lower upfront cost. Just be clear about that trade before you agree.
Can I get IP ownership even if I didn't pay for the full development?
Sometimes a vendor offers a model-as-a-service deal. The upfront cost is lower. But the IP is shared or kept by the vendor. This can work for an early test of an idea. Still, it creates long-term dependency. Do you plan to build your business on this AI? If so, negotiate full ownership upfront. Buying it back later usually costs more.
Sources
US Copyright Office - Copyright and artificial intelligence: ownership and authorship. copyright.gov
World Intellectual Property Organization - Emerging AI IP frameworks. wipo.int
Law.com - AI development contracts: key clauses and negotiating points. law.com
A clear ownership contract can give you the model weights, the code, and the pipelines. Let's talk about your project.
Book a free Brand and Tech Assessment to see exactly how we would grow your organic visibility.
Why Through The Glass Creatives
Knowing the strategy is the easy part. The hard part is running it at a level that moves your business. That is where most teams stall. That is the work Through The Glass Creatives does. TTGC is a premium brand, growth, and AI and development studio. The studio pairs elite brand thinking with hands-on technical work. That mix is rare. It is also what makes the team able to deliver work like this well. Book a free Brand and Growth Assessment to see how.
