Question 19 of 34
How do we handle intellectual property and copyright in AI?
Published by AI Governance Institute · Practical Governance for Enterprise AI
Navigating ownership of AI-generated content, copyright exposure from training data, and the contractual protections needed for AI-assisted work product.
If you only do 3 things, do this:
- 1.Audit AI vendor agreements for IP provisions: who owns outputs, does the vendor claim any license to outputs, and are you indemnified against copyright claims from training data?
- 2.If you're training models on third-party content, document the provenance and licensing of every training dataset before you start. Retroactive remediation is much harder.
- 3.Review client engagement agreements for AI use disclosure and IP allocation. Proactive disclosure is better than retroactive discovery.
The Situation
Who this is for: Legal, compliance, and procurement teams managing IP exposure from AI systems
When you need this: When procuring AI services, before initiating model training, or when reviewing client engagement agreements
The Decision
Who owns AI-generated outputs, are we exposed to copyright infringement claims from training data, and are our client agreements adequate?
The Steps
- 1Audit all AI vendor contracts for: output ownership, vendor license claims on outputs, copyright indemnification
- 2For internally trained models: document training data sources and licensing; identify any datasets using copyrighted content without clear license
- 3Assess whether AI-generated work product is copyright-protectable given current US Copyright Office guidance on human authorship
- 4Review client engagement agreements for AI use disclosure obligations and IP allocation
- 5For high-copyright-risk uses, negotiate vendor indemnification or seek explicit training data licenses
- 6Update standard contract templates to include AI-specific IP provisions
The Artifacts
- —AI vendor IP audit checklist (output ownership, indemnification, license claims)
- —Training data IP assessment template (source, license type, copyright risk rating)
- —Client engagement AI disclosure language template
- —IP clause library for AI contracts (output ownership, indemnification, disclosure)
The Output
A documented IP risk assessment for each AI vendor relationship and training dataset, updated contract templates with AI-specific provisions, and a client disclosure process in place.
AI-generated content and copyright ownership
Current U.S. copyright law requires human authorship for copyright protection. The Copyright Office has declined to register works produced autonomously by AI systems and has indicated that AI-generated content without sufficient human creative contribution is not eligible for copyright protection. This creates uncertainty about the IP status of AI-generated code, marketing materials, reports, and other work product.
The practical implication is that AI-generated content may be in the public domain, freely usable by competitors without any copyright claim. Organizations relying on AI-generated content for competitive advantage should evaluate whether that content is actually protectable. Where human creative contribution is substantial and documented, copyright may attach to the human-authored portions.
Training data and copyright exposure
Several major AI developers are defendants in copyright litigation alleging that training models on copyrighted content without license constitutes infringement. The legal outcome of these cases remains uncertain, but the risk is material for organizations training their own models on internet-sourced or third-party content.
If you are training or fine-tuning models, document the provenance of your training data and the legal basis for its use. Prefer datasets that are explicitly licensed for AI training use. For third-party models you deploy, review vendor representations about training data provenance and indemnification coverage for copyright claims arising from model outputs.
Contractual protections
Review your AI vendor agreements for IP provisions covering: who owns outputs generated using the service, whether the vendor claims any license to outputs, and whether the vendor indemnifies you for copyright claims arising from the model's training data or outputs. Vendor positions vary significantly on these points and are negotiable in enterprise agreements.
For AI-assisted work product delivered to clients, review your engagement agreements to ensure they accurately reflect the use of AI tools and allocate IP and liability appropriately. Clients in regulated industries may have their own policies about AI use by outside counsel, consultants, and vendors. Proactive disclosure is generally preferable to retroactive discovery.