IP, Training Data & Open Source
Who owns the model, the output and the code. Where the training data came from and under what terms. Which open-source licenses are in your stack and what they require.
These are the first questions an investor or acquirer asks an AI company. We help you have the answers, and the documents behind them, before anyone asks.
An AI company’s value is its IP. Its IP is only what it can prove.
The Copyright Office has said that purely machine-generated output is not protected by copyright, while human contributions can be. Ownership between you and your model provider is set by the provider’s terms. Training data carries whatever license it came with.
None of that is a problem until a financing, an acquisition or a claim. Then every gap becomes a price adjustment or an indemnity.
We build the record as you go: assignments from everyone who writes code, a provenance log for training data, a license inventory for the stack, and copyright and trade secret protection for what matters most.
The lines the IP statutes draw.
Figures from the Copyright Act, the Patent Act and the Defend Trade Secrets Act.
Under 17 U.S.C. § 411, registration generally must be made before an infringement suit can be filed. Statutory damages run from $750 to $30,000 per work before the willfulness increase. Sources: 17 U.S.C. § 504, Copyright Office; 17 U.S.C. § 411, Copyright Office; USPTO, MPEP § 2153, AIA 35 U.S.C. 102(b)(1); 18 U.S.C. § 1836, Defend Trade Secrets Act; U.S. Copyright Office, Copyright and Artificial Intelligence.
What we do on IP for technology companies.
Assignments and chain of title
Founder, employee and contractor IP assignments, confirmatory assignments for past work, and the records investors ask for in diligence.
Provenance and licensing
Where the data came from, under what license, what it permits, and the agreements for data you license in or out.
Model and output rights
What your model provider’s terms say about inputs and outputs, what you can claim as your own, and what you pass through to your users.
Copyright and trade secrets
Registering the code and content that matter, and the access controls, confidentiality terms and policies that make a trade secret a trade secret.
License compliance
An inventory of the open-source components in your stack, what each license requires, and remediation before a financing or sale finds the problem.
Patent strategy and coordination
When a patent filing makes sense for an AI or software invention, the one-year clock after public disclosure, and coordination with patent counsel.
How an IP review runs.
Done once properly, then maintained as the product grows.
Inventory
Who wrote what, what data trained what, and what licenses govern what.
Gaps
Missing assignments, unlicensed data, incompatible open-source terms and unregistered works that matter.
Cure
Confirmatory assignments, license negotiations, component replacement and registrations.
Record
A diligence-ready IP file and templates that keep it current.
We draft it. We also defend it.
Fintech and digital assets since 2017.
Max Dilendorf has advised fintech, payments and digital-asset companies since 2017, through the growth of the sector and the regulation that followed. The questions founders ask today about AI, we have worked through before with crypto.
More than 130 arbitrations since 2019.
Software liability, platform and payment disputes, cybercrime losses and AI-related claims before the AAA, JAMS and NAM, and emergency relief in New York courts. What we learn in those rooms shapes how we draft.
Fix the chain of title before someone else checks it.
Every gap found in diligence costs more than the same gap fixed earlier.
- Get a signed IP assignment from every founder, employee and contractor
- Keep a log of training data sources and the license for each
- Read your model provider’s terms on output ownership before you rely on them
- Run an open-source license scan and keep the inventory current
- Register the copyright in your core code and content before you need to sue
- Assume contractor work belongs to you without a written assignment
- Train on data you do not have the right to use because it was publicly available
- Describe AI-generated output as proprietary without knowing what you can claim
- Ship a product on copyleft code without understanding the obligations
- Disclose an invention publicly and wait more than a year to file
From the firm’s writing.
AI Startup Attorney: Legal Guide for Founders & Seed Funding
The documentation investors ask AI founders for, including IP and data.
Read the articleDomain Account Takeover: What Victims Must Do in the First 48 Hours
When a company’s domain, the front door to its brand, is taken.
Read the articleAsset Protection Lawyer for Tech Founders & Crypto Trusts
How founders separate personal wealth from company risk.
Read the articleFrequently asked.
Between you and the provider, the provider’s terms decide, and they vary. As against the world, the Copyright Office’s position is that purely machine-generated output is not protected by copyright, while human-authored contributions can be. We document the human contribution and protect the rest through contracts and trade secret measures.
Publicly available is not the same as licensed. Websites have terms, datasets have licenses, and copyrighted works are copyrighted whether or not they are online. The Copyright Office’s third report addresses training on copyrighted works. We help you document what you used, under what terms, and where the risk is.
Not automatically. Unless the contractor signed a written assignment, the contractor may own the copyright and you may hold only a license. A confirmatory assignment can cure it if the contractor cooperates. Every contractor agreement should include one from the start.
Not open source itself; most of your stack is open source. The problem is license terms you did not read. Some licenses require you to release your own source code if you modify and distribute, or offer over a network. Acquirers and investors scan for this. We inventory and remediate before they do.
Sometimes. Software and AI inventions can be patentable, but the cost and disclosure must be weighed against trade secret protection. One clock matters regardless: a public disclosure starts a one-year period to file in the United States. We help decide and coordinate with patent counsel.
Other areas of the practice.
Six services for technology companies, from ongoing counsel to the dispute. Back to the overview.
Know what you own before someone asks you to prove it.
A confidential review of your IP chain of title, training data and open-source stack, with a diligence-ready file at the end.