AI and Intellectual Property: Legal Guide for Enterprises

AI and intellectual property law are colliding in ways that create genuine enterprise risk. Companies are using AI to create content, train models on proprietary data, and produce outputs that may infringe third-party rights. Meanwhile, they are investing in AI capabilities they want to protect as competitive assets.

This guide provides the practical IP framework that enterprise legal and technology teams need.

IP Issues in AI: The Four Key Questions

Can AI-generated outputs be owned, and by whom?
Does training an AI on copyrighted material infringe copyright?
What IP risks exist when using third-party AI tools with proprietary data?
How can enterprises protect their AI investments as IP assets?

Question 1: Ownership of AI-Generated Outputs

The current state of law (2026):

In the United States, the Copyright Office has consistently held that copyright requires human authorship. Works created entirely by AI with minimal human creative contribution do not receive copyright protection.

Works where AI is a tool used by a human creator — where the human makes meaningful creative choices — can receive copyright protection, but the scope is limited to the human-created elements.

The practical implication: If your marketing team uses AI to generate the first draft of an article, and human editors make significant creative revisions, the final work is likely copyrightable. If AI generates a full article with no meaningful human creative contribution, it is not.

In the EU: Similar principles — copyright requires human creative authorship. The EU AI Act does not address copyright directly; existing copyright law applies.

Enterprise recommendation: For AI-generated content you want to protect, document the human creative contributions. Maintain records of human review, selection, and creative modification of AI outputs.

Question 2: Training Data and Copyright

The debate: Does training AI models on copyrighted works constitute infringement?

This is currently the central AI copyright litigation question, with multiple cases pending in the US (Getty Images v. Stability AI, The New York Times v. OpenAI, class actions by authors and artists).

The competing arguments:

For infringement: Copying copyrighted works to train AI models is reproduction without license. The outputs that reproduce substantial portions of training data (demonstrable in some image generators) are derivative works requiring permission.

Against infringement: Training is transformative use (fair use under US law). The AI doesn't reproduce the work; it learns statistical patterns from it.

Current state: US courts have not definitively resolved this. Until they do, enterprises face meaningful legal uncertainty.

Enterprise recommendations:

For training your own models: prefer datasets with clear licenses (public domain, Creative Commons, commercial licenses that explicitly permit training use)
Audit data sources for any training initiatives
Keep records of training data composition and licensing

Question 3: IP Risks When Using Third-Party AI Tools

When enterprises use commercial AI tools (OpenAI, Anthropic, Google, etc.) with proprietary data, several IP risks emerge:

Trade secret contamination: If you input proprietary business information into a third-party AI API that uses inputs for training, that information may end up in the model and potentially be surfaced to other users.

Mitigation: Review AI provider data usage policies carefully. Most enterprise tiers (OpenAI Enterprise, AWS Bedrock, Azure OpenAI) explicitly commit to not using customer data for training. Verify this contractually.

Output indemnification: Some AI providers have offered contractual indemnification for IP claims arising from their AI outputs. Microsoft Copilot Copyright Commitment and Google's generative AI IP indemnification are examples. Understand what protection your AI vendor provides.

License incompatibility: Some open-source AI models have licenses that restrict commercial use or require derivative works to be shared under the same terms. Verify license compatibility before deploying open-source models in commercial products.

Competitive intelligence risk: Employees submitting proprietary strategy documents, financial data, or customer lists to AI tools may be inadvertently disclosing trade secrets. This is an enterprise data governance issue as much as a legal one.

Question 4: Protecting AI Investments as IP Assets

Enterprises investing significantly in AI should think about how to protect those investments:

Trade secrets: The most practical IP protection for AI systems. AI model weights, training datasets, system architectures, and prompts can all qualify as trade secrets if:

They provide competitive advantage
They are kept confidential
Reasonable steps are taken to maintain confidentiality

Practical requirements: Access controls, NDAs with employees and contractors, confidentiality provisions in contracts, documented trade secret policies.

Patents: AI-related patents are possible for novel technical inventions — new training methods, new architectures, new applications of AI to specific problems. Pure AI models are not patentable as such, but methods and systems using AI may be.

Important limitation: Patent protection for AI is complex and jurisdiction-dependent. Consult IP counsel before filing.

Copyright in training datasets: Curated datasets can be protected as compilations (selection and arrangement of data) even if individual elements are not copyrightable. Proprietary training datasets should be documented and access-controlled.

Building an Enterprise AI IP Policy

An effective AI IP policy addresses:

Use of third-party AI tools: Which tools are approved? What data can be input into them? What are the restrictions?
AI-generated content ownership: How does the company claim ownership of AI-assisted work? What human contribution requirements apply?
Training data governance: What data can be used to train proprietary AI models? What licensing review is required?
Trade secret protection: What AI assets are trade secrets? How are they identified, labeled, and protected?
Output IP review: When AI is used to produce customer-deliverable content, what IP review is required before delivery?

Key Contract Provisions for AI

Ensure your AI vendor contracts address:

Data usage: Explicit commitment that input data will not be used for model training
Output ownership: Confirmation that outputs generated from your inputs belong to you
IP indemnification: Coverage for IP claims arising from AI outputs
Data deletion: Right to have input data deleted upon request or contract termination
Confidentiality: Appropriate protections for proprietary data shared with the vendor

Conclusion

AI and IP law is in a dynamic and unsettled state. The legal framework will continue to evolve as courts and legislators address the novel questions AI raises. Enterprises that build proactive IP policies now — governing training data, protecting AI assets, and managing output IP — are better positioned to navigate this uncertainty than those that wait for the law to fully clarify.