Who owns the IP and code when you hire an AI agency?
Who owns the code and IP when you hire an AI agency depends entirely on the contract — ownership does not transfer automatically just because you paid. To own the source code, the fine-tuned model, and any derivative work, you need an explicit IP assignment in the Statement of Work, agreed before work starts. AI adds extra layers — the model, the training data, and third-party components — each of which must be named.
IP disputes surface at the worst possible moment: when the project has succeeded and the system is most valuable. The fix is cheap and entirely upfront — name every layer in the contract before anyone writes code.
The source code
The application and integration code written for your project. This should be assigned to you in writing. Without an explicit assignment, the agency may retain rights even though you paid for it.
The fine-tuned model
If the agency fine-tunes a model on your data, the resulting weights are valuable and distinct from the base model. Specify who owns them. This is the layer buyers most often forget, and the one that hurts most to lose.
The training data
Your data is yours, but the processed datasets and derived artefacts created during training are a grey area unless named. State that derived data belongs to you and isn't reused elsewhere.
Third-party base models and components
Most AI systems build on licensed foundation models and open-source libraries you can't own outright. The agency should disclose these and their licence terms, so you understand any ongoing obligations or restrictions.
Everything you need to run it
Ownership is worthless if you can't operate the system. Require handover of the code, model weights, configuration, and documentation on completion — not access that depends on the agency staying in the picture.
Where ownership gets settled
IP belongs in your Statement of Work, agreed before work begins. It's one of the specific things to confirm when you ask an agency the right questions, and vagueness here is a red flag.
Ownership, clear from the start
NorthBridge AIengagements run on a signed Statement of Work where code and IP ownership are set out before the project begins, so you're never negotiating ownership after the fact. See how it works.
Frequently asked questions
Who owns the code when you hire an AI agency?
It depends entirely on the contract. Unless the agreement explicitly assigns ownership to you, the agency may retain rights to the code it writes. To own the source code, the fine-tuned models, and any derivative work, you need a written IP assignment clause in the Statement of Work, agreed before the project starts. Never assume ownership transfers automatically because you paid for the work.
Do I own the AI model an agency builds for me?
Only if the contract says so. A model has several layers — the underlying base model (often licensed from a third party), any fine-tuning, the training data, and the surrounding code. Your contract should be explicit about which of these you own or license, especially the fine-tuned weights and the data used to produce them. Third-party base models will carry their own licence terms you also need to understand.
What IP clauses should be in an AI agency contract?
The contract should assign ownership of the source code and derivative work to you, clarify ownership or licensing of any fine-tuned models and training data, disclose third-party components and their licences, and confirm you receive everything needed to run and maintain the system (code, weights, documentation) on completion. Settle all of this before work begins.
Own what you pay for
Code and IP ownership defined up front on every NorthBridge AI project.
Browse verified agencies