How Can an Intellectual Property Strategy Be Built for AI-Based Products?
An intellectual property strategy for an AI-based product goes beyond asking who owns the copyright in its outputs. Datasets, software code, third-party models, APIs, prompt structures, model…
What’s in this article?
An intellectual property strategy for an AI-based product goes beyond asking who owns the copyright in its outputs. Datasets, software code, third-party models, APIs, prompt structures, model configurations, brand assets and employee-created content must be considered together.
Therefore, intellectual property management for AI products is a governance responsibility from the start of architectural design, not merely a legal check at the end of development.
Which assets should an AI product protect?
- Source code and custom software components,
- datasets used for training or fine-tuning,
- organisation-specific prompts and system instructions,
- RAG knowledge bases and data-preparation methods,
- model configurations and workflows,
- generated text, images, audio and other outputs,
- product name, logo and other brand assets,
- methods and operational information that constitute trade secrets.
1. Review the licences of the models and APIs you use
For a product using a third-party model or API, first review the provider’s contractual and licence terms. Commercial use, use of outputs, data retention, use of data for model development and liability terms can vary between providers.
Therefore, the model chosen by the technical team should be evaluated not only in terms of performance but also in terms of legal and commercial conditions of use.
2. Document the source of training data
“Found online” does not identify a data source. Record where each dataset comes from, its licence, whether it contains personal data and whether it includes material protected by third-party intellectual property rights.
Recording the provenance of corporate datasets makes potential rights and licensing disputes easier to manage.
3. Clarify Employee and Supplier Contracts
Employees, freelancers or external software companies may develop the code, data-preparation processes or model optimisations. Contracts must clearly define intellectual property rights, permitted uses, confidentiality and deliverables.
Receiving a technical deliverable does not, by itself, mean that all necessary rights automatically transfer to the company.
4. Do not base ownership of AI outputs on a single assumption
Protection and ownership of AI-generated content may depend on human contributions, the system used, contractual terms and the applicable legal system. Companies should therefore develop policies for specific use cases rather than assume that “AI created it, so all rights belong to us”.
5. Treat brand protection as a separate layer
As an AI product’s name, logo and identity gain prominence, its trademark strategy becomes more important. A technically strong model does not prevent third parties from imitating the product name, domains or digital identity.
When determining the product name, brand research, domain strategy, and digital imitation risks should be evaluated together.
6. Keep trade secrets out of uncontrolled model inputs
Entering source code, customer information, pricing models or unreleased product plans into uncontrolled AI tools is a significant intellectual property risk.
Corporate AI policies must define which data may be used in each tool and which data must never be sent to external systems.
7. Establish output validation and similarity checks
Human review is needed especially for marketing materials, designs, software code and content created for clients. Depending on the risk of the intended use, outputs can be checked for problematic similarities to third-party protected materials.
Example: AI Based SaaS Product
If a company is developing AI-backed SaaS that analyzes its customers' documents, several layers of intellectual property are formed at once:
- The source code for the platform may be owned by the company.
- The basic model may be subject to a third party license.
- Ownership of customer documents may remain with the customer.
- The organisation-specific RAG and data-preparation system may constitute a trade secret.
- The terms of use of AI outputs can also be regulated in the customer contract.
These elements cannot be resolved by a single concept of "AI use contract".
AI intellectual property checklist
- Are the licences for the models and datasets in use documented?
- Is the source of the datasets known?
- Are intellectual property clauses in employee and supplier contracts up to date?
- Is the transfer of confidential information to AI tools restricted?
- Are product name and digital brand assets protected?
- Is there a human verification process for AI outputs?
- Is the rights chain being reevaluated when the model or provider is changed?
Conclusion
A strong intellectual property strategy for AI products requires coordination between legal, product, data and technical architecture teams. The aim is not to slow innovation, but to establish a sound chain of rights and licences for a scalable product from the outset.
B10’s services in Artificial Intelligence & Automation, Brand Protection & Monitoring and Custom Software & Mobile Applications help design these technical and digital layers together.
Sources and related documents
These documents provide references for the methods and rules discussed in this article.