Intellectual Property for AI & Technology Companies
Companies building with AI face IP questions that did not exist a decade ago. We help you protect models, navigate training-data issues, and structure an IP position investors can underwrite.
Protecting AI Models: Trade Secret or Patent?
For most AI companies, the model is the product — and deciding how to protect it is the foundational IP question. Patents and trade secrets pull in opposite directions. A patent requires public disclosure of how the invention works in exchange for a time-limited right to exclude others; a trade secret requires the opposite — keeping the information confidential — and lasts as long as the secret holds.
The right answer depends on observability. Innovations a competitor could reverse-engineer or independently observe from your product's behavior are weak trade secret candidates and may warrant patent protection. Innovations that stay behind your API — training procedures, data curation pipelines, model weights, internal optimizations — are often better held as trade secrets, provided you maintain the access controls, confidentiality agreements, and internal policies that trade secret law requires. Most companies end up with a deliberate mix, and we help you draw that line invention by invention rather than by default.
Training Data and Copyright
Whether training a model on copyrighted works constitutes infringement — and whether fair use applies — is among the most actively litigated questions in technology law, with cases pending against major AI developers and the law still developing. Companies building or fine-tuning models cannot wait for final answers; they need practical risk management now.
We counsel clients on the copyright posture of their training pipelines: documenting data provenance, evaluating dataset licenses, structuring agreements with data vendors, and assessing exposure for both model training and model outputs. We also address the flip side — what rights you hold in content your models generate, and how to handle copyright questions in customer contracts, since enterprise buyers increasingly ask for representations about training data and indemnification for AI outputs.
AI Inventorship and USPTO Guidance
U.S. patent law requires a human inventor. The Federal Circuit confirmed in Thaler v. Vidal that an AI system cannot be named as an inventor, and the USPTO's 2024 inventorship guidance addresses the harder, more common question: when is an invention developed with AI assistance still patentable? Under that guidance, AI-assisted inventions remain eligible for patenting when a natural person made a significant contribution to the invention — but each claim must be supported by such a contribution, and merely presenting a problem to an AI system or recognizing its output as useful may not suffice.
For engineering teams that use AI tools daily, this has practical consequences. We help clients document the human contribution in their invention process, structure invention disclosure practices that capture who contributed what, and evaluate inventorship on a claim-by-claim basis before filing — because inventorship errors can put an issued patent at risk.
IP Due Diligence for Acquisitions and Investment
When a technology company is acquired or raises a significant round, its IP position gets examined. Diligence routinely surfaces the same issues: contractors who never assigned their IP, open-source components with obligations incompatible with the company's licensing model, unregistered marks, gaps in confidentiality practices, and patent applications that lapsed for missed deadlines. Found late, these issues cost leverage; found early, most are fixable.
We perform IP due diligence from both sides of the table — for acquirers and investors evaluating a target, and for companies preparing to be evaluated. For sell-side clients, a pre-diligence audit identifies and cures problems before a counterparty finds them. For buy-side clients, we translate the IP findings into deal terms: representations, indemnities, escrows, and price.
How We Help
- AI-generated content ownership
- Training data copyright implications
- Model protection strategies (trade secret vs. patent)
- USPTO AI inventorship guidance
- IP due diligence for software acquisitions
- IP portfolio development for tech companies
Frequently Asked Questions
Can an AI system be named as an inventor on a patent?
No. U.S. law requires a natural person as the inventor, and the Federal Circuit confirmed in Thaler v. Vidal that an AI system cannot be one. However, inventions developed with AI assistance can still be patented under the USPTO's 2024 guidance — provided a human made a significant contribution to each claimed invention. Documenting that human contribution is the key practical step.
Should we patent our model or keep it a trade secret?
It depends on what a competitor can observe. If the innovation is visible or reverse-engineerable from your product, trade secret protection is fragile and a patent may be worth the disclosure. If it stays behind your API — training methods, data pipelines, model weights — trade secret protection often makes more sense, as long as you maintain real confidentiality measures. Most AI companies protect different layers of the stack differently.
Is it legal to train a model on copyrighted data?
The law is unsettled. Whether training on copyrighted works is fair use is being actively litigated, and outcomes may vary by how the data was obtained, what the model does, and how outputs compare to the source works. In the meantime, companies can manage risk through data provenance records, licensed datasets, vendor agreements, and careful treatment of model outputs — which is exactly the posture sophisticated customers and investors now expect.
Ready to Protect Your Innovation?
Schedule a confidential consultation to discuss your intellectual property needs.