AI and Intellectual Property: Navigate Copyright and Ownership

The Intellectual Property Crisis AI Has Created

Artificial intelligence has fundamentally disrupted the intellectual property frameworks that have governed innovation and creativity for centuries. These frameworks were built on a simple premise: humans create, and the law protects their creations. AI shatters this premise by introducing a non-human entity that can generate text, images, music, code, designs, and inventions at unprecedented scale and speed.

The resulting legal uncertainty is enormous and carries real business consequences. Can AI-generated content be copyrighted? Who owns the output of an AI model trained on millions of copyrighted works? Can an AI system be named as an inventor on a patent? If your AI model generates content that closely resembles someone else's copyrighted work, who is liable?

These questions are not abstract. In 2025, the value of AI-generated content exceeded $50 billion globally, and disputes over AI intellectual property multiplied accordingly. The US Copyright Office received over 12,000 applications involving AI-generated or AI-assisted works in 2025, grappling with each on a case-by-case basis. Major lawsuits including The New York Times v. OpenAI, Getty Images v. Stability AI, and multiple class-action suits from authors and artists are reshaping the legal landscape in real time.

For enterprises deploying AI at scale, understanding and navigating AI intellectual property issues is not optional. It determines whether your AI investments create protectable assets or unprotectable commodities, and whether they generate legal liability or legal advantage.

Copyright and AI-Generated Content

The intersection of copyright law and AI-generated content is the most actively contested area of AI intellectual property law.

The Fundamental Question: Can AI Outputs Be Copyrighted?

Copyright law in most jurisdictions requires human authorship. Works created entirely by machines, animals, or natural processes have traditionally been uncopyrightable. The US Copyright Office's 2023 guidance confirmed this principle: purely AI-generated content, where a human merely prompts an AI and publishes the output, is not eligible for copyright protection.

However, the situation becomes more nuanced when humans are significantly involved in the creative process. The Copyright Office has recognized protection for works where humans make creative decisions about how to use AI, including selecting, arranging, and modifying AI outputs. The key question is the degree of human creative control.

The Copyright Office's 2025 rule on AI-assisted works established a spectrum:

• **Fully AI-generated**: No copyright protection. A prompt like "generate a landscape painting" produces an uncopyrightable output.
• **AI-assisted with significant human creativity**: Copyrightable, but only the human-authored elements are protected. An artist who uses AI to generate base images, then substantially modifies, arranges, and combines them with original elements, can claim copyright in the resulting work.
• **Human-created with AI tools**: Fully copyrightable. Using AI as a tool (like spell-check or Photoshop's content-aware fill) does not diminish copyright in an otherwise human-created work.

Implications for Enterprise Content

For businesses generating content with AI, the practical implications are significant:

• **Marketing content**: If your marketing team uses AI to generate copy, images, or video, the purely AI-generated portions may not be copyrightable. Competitors could legally copy AI-generated marketing materials without infringement. To strengthen protection, ensure human creative involvement in selection, arrangement, and modification.
• **Software code**: AI-generated code exists in a legal gray area. The Copyright Office has not issued definitive guidance on AI-generated code specifically, but the general principle of human authorship applies. Code that is substantially modified by human developers has stronger copyright claims than code used verbatim from AI output.
• **Training data and documentation**: AI-generated documentation, reports, and training materials may lack copyright protection. Organizations should document the human creative contributions to these works to support potential copyright claims.

The Training Data Copyright Controversy

The other side of the copyright equation concerns the data used to train AI models. Most large AI models are trained on vast datasets that include copyrighted works, often without the explicit consent of copyright holders.

The legal question is whether using copyrighted works to train AI models constitutes fair use (in the US) or falls under existing copyright exceptions (in other jurisdictions). This question is being litigated in multiple high-profile cases:

• **The New York Times v. OpenAI**: The Times argues that training on its articles constitutes infringement and that ChatGPT can reproduce near-verbatim excerpts. OpenAI argues fair use.
• **Getty Images v. Stability AI**: Getty argues that using its images to train Stable Diffusion infringes its copyrights. Stability AI argues fair use and transformation.
• **Authors Guild v. OpenAI**: A class action on behalf of authors whose books were used in training data.

The outcomes of these cases will fundamentally shape the AI industry. If training on copyrighted data is found to be infringing, AI companies may face massive liability and need to license training data or rely on public domain and explicitly licensed content. If fair use prevails, the current model of training on internet-scale data will continue, but with potential obligations around output filtering and attribution.

For enterprises using third-party AI models, the risk of downstream copyright liability is a key consideration. If a model was trained on infringing data, are you liable for using its outputs? Most AI providers include indemnification clauses in their enterprise agreements, but the scope and enforceability of these provisions are untested. For a broader view of AI compliance challenges, see our guide on [AI compliance in regulated industries](/blog/ai-compliance-regulated-industries).

Patents and AI Inventions

AI's impact on patent law raises two distinct questions: can AI be named as an inventor, and can AI-related innovations be patented?

AI as Inventor

The DABUS case, where researcher Stephen Thaler sought patents naming his AI system as the sole inventor, tested patent offices worldwide. The unanimous answer so far has been no. The US Patent and Trademark Office, the European Patent Office, the UK Intellectual Property Office, and courts in Australia and South Africa (after an initial reversal) have all held that patent law requires a human inventor.

In February 2024, the USPTO issued guidance clarifying that while AI cannot be an inventor, humans who use AI in the inventive process can be named as inventors, provided they made a "significant contribution" to the conception of the invention. Merely instructing an AI to solve a problem is not sufficient. The human must contribute intellectually to the inventive concept.

For enterprises, this means:

• **Document human contributions**: When AI assists in the inventive process, carefully document the human intellectual contributions at every stage. Who identified the problem? Who designed the approach? Who evaluated and selected among AI-generated options? Who refined the solution?
• **Inventor identification**: Ensure that the named inventors on patent applications genuinely contributed to the invention's conception, not just the AI's operation.
• **Trade secret consideration**: For innovations where documenting sufficient human inventorship is difficult, trade secret protection may be more appropriate than patents.

Patentability of AI Innovations

AI-related inventions face specific patentability challenges:

• **Abstract idea rejection**: In the US, under the Alice/Mayo framework, many AI innovations are rejected as abstract ideas implemented on a computer. To overcome this, patent applications must demonstrate a specific, practical application that improves a technical process, not just a general-purpose AI method.
• **Obviousness concerns**: As AI techniques become widely known, examiners increasingly reject applications that apply known AI methods to new domains as obvious. Demonstrating non-obvious improvements in accuracy, efficiency, or capability is essential.
• **Disclosure requirements**: Patent applications must disclose enough detail for a skilled practitioner to reproduce the invention. For AI systems, this may require disclosing model architectures, training methodologies, and sometimes training data characteristics, creating tension with trade secret protection.

Despite these challenges, AI-related patents continue to be granted in large numbers. In 2025, the USPTO granted over 40,000 patents related to AI and machine learning, primarily for specific applications and improvements rather than fundamental algorithms.

Trade Secrets and AI

Trade secret protection is often the most practical IP strategy for AI assets. Unlike patents, trade secrets do not require disclosure, can protect information that is not patentable (such as training data and hyperparameters), and last indefinitely as long as secrecy is maintained.

What AI Assets Can Be Trade Secrets?

Virtually any confidential information that provides competitive advantage can qualify:

• **Training data**: Curated, labeled datasets that provide competitive advantage.
• **Model architectures**: Custom architectures developed for specific applications.
• **Hyperparameters and training configurations**: The specific settings that produce optimal results.
• **Feature engineering**: Novel features and preprocessing pipelines.
• **Evaluation methodologies**: Proprietary methods for evaluating model quality.
• **Data pipelines**: Efficient systems for collecting, processing, and managing training data.

Protecting AI Trade Secrets

Trade secret protection requires reasonable measures to maintain secrecy:

• **Access controls**: Limit access to AI assets to employees and contractors who need them, and implement technical controls (encryption, access logging, DRM) to enforce restrictions.
• **Confidentiality agreements**: Ensure all employees, contractors, and partners who access AI assets are bound by confidentiality obligations.
• **Departure procedures**: Implement procedures to protect trade secrets when employees leave, including exit interviews, equipment retrieval, and access revocation.
• **Documentation**: Maintain records of what constitutes trade secrets, what protective measures are in place, and who has access.

The Girard AI platform provides comprehensive access controls and audit logging that support trade secret protection for AI assets. For detailed audit and logging strategies, see our guide on [AI audit logging and compliance](/blog/ai-audit-logging-compliance).

Practical IP Strategy for AI-Driven Enterprises

Audit Your AI IP Portfolio

Start by inventorying all AI-related intellectual property across your organization:

• **Models**: What models have you developed, and what is unique about them?
• **Data**: What proprietary datasets have you created, and how are they protected?
• **Code**: What custom AI code, frameworks, and tools have you built?
• **Processes**: What novel AI development processes, evaluation methods, or deployment techniques have you developed?
• **Outputs**: What AI-generated content and materials are you producing, and what are the copyright implications?

Determine the Right Protection Strategy

For each asset, evaluate the optimal combination of protection mechanisms:

| Asset Type | Patent | Trade Secret | Copyright | Contract | |-----------|--------|-------------|-----------|----------| | Novel algorithms | Strong candidate | Good backup | Limited | Supplement | | Training data | Rarely applicable | Primary protection | Database rights may apply | Essential | | Model weights | Rarely applicable | Primary protection | Uncertain | Essential | | AI-generated content | Not applicable | Limited | Depends on human involvement | Important | | Custom tools/code | Possible | Good option | Yes, for code | Supplement |

Manage Third-Party AI IP Risks

When using third-party AI models and services, conduct due diligence on IP risks:

• **Review license terms**: Understand what rights you have to outputs generated by third-party models. Can you use them commercially? Can you claim ownership? Are there restrictions on how outputs can be used?
• **Evaluate training data provenance**: Assess the risk that the model was trained on data that creates downstream copyright liability for you.
• **Secure indemnification**: Negotiate indemnification clauses that protect you if the AI provider's model generates outputs that infringe third-party IP.
• **Monitor for infringement**: Implement systems to check AI outputs for similarity to known copyrighted works before publishing or distributing them.

Develop an AI IP Policy

Create a comprehensive organizational policy that addresses:

• **Ownership of AI-assisted inventions**: How inventorship is determined when AI assists in the creative or inventive process.
• **AI-generated content guidelines**: What level of human involvement is required before content can be published under the organization's name.
• **Third-party AI usage rules**: What third-party AI tools employees can use and what IP considerations apply.
• **Disclosure obligations**: When and how employees must disclose the use of AI in creating work product.
• **Training data policies**: How copyrighted materials can and cannot be used in training proprietary AI models.

For governance frameworks that include IP considerations, review our guide on [AI governance framework best practices](/blog/ai-governance-framework-best-practices).

Emerging Trends and Future Outlook

Legislative Developments

Several jurisdictions are developing AI-specific IP legislation:

• **EU AI Act and Copyright Directive**: The AI Act's interaction with the Copyright Directive creates specific obligations for AI providers regarding training data transparency and opt-out mechanisms for copyright holders.
• **US Congressional activity**: Multiple bills addressing AI and copyright are under consideration, including proposals for mandatory training data disclosure, AI output labeling, and compulsory licensing frameworks.
• **International harmonization**: WIPO is facilitating discussions on harmonizing AI IP treatment across jurisdictions, though consensus remains elusive.

Industry Self-Regulation

In the absence of settled law, industry practices are emerging as de facto standards:

• **Content authenticity initiatives**: The C2PA standard for content provenance is being widely adopted.
• **Training data licensing**: Several marketplaces for licensed AI training data have emerged, providing a legal alternative to scraping.
• **Output attribution**: Some AI providers are implementing attribution systems that credit training data sources when outputs are closely derived from specific works.

Technology Solutions

Technical solutions are emerging to address IP challenges:

• **Provenance tracking**: Systems that track the origin and modification history of AI-generated content.
• **Similarity detection**: Tools that compare AI outputs against databases of copyrighted works to flag potential infringement.
• **Unlearning techniques**: Methods for removing specific copyrighted works from trained models without full retraining.

Protect Your AI Investments

AI intellectual property is a complex, rapidly evolving area where the legal frameworks are still being written. Organizations that proactively develop AI IP strategies will protect their investments, reduce legal risk, and position themselves to capture the full value of their AI innovations.

Do not wait for the law to settle. The decisions you make today about how you develop, protect, and deploy AI will determine your IP position for years to come. Audit your AI IP portfolio, implement appropriate protections, manage third-party risks, and stay engaged with the evolving legal landscape.

[Contact our team](/contact-sales) to learn how the Girard AI platform helps enterprises manage AI intellectual property with comprehensive access controls, audit logging, and provenance tracking, or [sign up](/sign-up) to explore our IP management features.