New Guidance on AI-Related Inventions: What Innovators Need to Know

Always document the inventors’ role in AI-assisted discoveries to avoid issues with inventorship.

Introduction

Artificial Intelligence (AI) is transforming industries at an unprecedented pace, from healthcare and finance to logistics and entertainment. With this rapid evolution comes a surge in AI-related intellectual property (IP) challenges, spanning patents, trademarks, and copyrights. In response, the U.S. Patent and Trademark Office (USPTO) and other global IP offices have issued new guidance to clarify how AI-assisted innovations are examined and protected.

This article unpacks the recent guidance and explores its implications for inventors, companies, and IP practitioners. Whether you’re an AI startup, an established technology company, or an investor in AI-driven businesses, understanding these developments is critical to protecting and monetizing your intellectual property.

AI and Patentability: The Core Issues

The USPTO's guidance focuses on three key areas for AI-related patents:

1. Inventorship in AI-Generated Inventions

2. Patent Eligibility Under Existing Legal Frameworks

3. Disclosure and Enablement Requirements

1. Inventorship in AI-Generated Inventions

One of the most pressing legal issues in AI patents is who (or what) can be named as an inventor. The USPTO has made it clear: only human inventors qualify. This position aligns with a series of legal battles over AI-generated inventions, such as the well-known Thaler v. Vidal case. U.S. courts ruled that an AI system, even one that autonomously generates novel and useful ideas, cannot be listed as an inventor. A human must have made a significant contribution to the inventive process.

Practically speaking, companies must document the role of human inventors in AI-generated innovations. If an AI system merely executes pre-programmed tasks, its human operators can claim inventorship. However, if AI autonomously generates an invention, securing patent rights becomes more complex.

2. Patent Eligibility: When is an AI-Related Invention Patentable?

The USPTO’s guidance confirms that AI-related inventions must meet the same subject matter eligibility requirements under 35 U.S.C. § 101 as any other technology. To pass the Alice/Mayo test, an AI invention must:

• Not be an abstract idea (e.g., merely an algorithm or mathematical model), or

• If abstract, include an inventive concept that transforms it into a patentable application.

For example, an AI model that simply predicts stock prices based on historical data may be deemed an unpatentable abstract idea. However, an AI system that improves semiconductor manufacturing efficiency by uniquely processing sensor data could qualify. The key is showing that the AI’s application leads to a real-world technological improvement.

3. Disclosure and Enablement: The New “Black Box” Problem

A growing challenge in AI patents is the enablement requirement under 35 U.S.C. § 112. Many AI models, particularly deep learning networks, operate as "black boxes"—even their creators struggle to explain how they arrive at certain outputs. Patent law, however, demands that an invention be described with sufficient detail so that a person skilled in the art can reproduce it.

The new guidance emphasizes that AI patents must include:

• A clear and complete disclosure of how the AI system functions.

• Details on training data and how the AI reaches its conclusions.

• Sufficient explanation to ensure a skilled person could implement the invention without undue experimentation.

Simply put, applicants cannot claim broad AI-based inventions without revealing the specifics of how their AI works.

Trademarks and Copyright in the AI Era

Trademarks and AI

AI is increasingly being used in branding and marketing, raising unique trademark issues:

• AI-Generated Trademarks: Similar to patents, trademarks created by AI systems cannot be registered without a human applicant. The USPTO requires that a human entity be responsible for the application.

• Likelihood of Confusion: AI tools used to generate brand names or logos must ensure they do not infringe on existing trademarks. Errors in training data could lead to unintentional infringement.

• Recent Case: In Booking.com B.V. v. USPTO, the Supreme Court clarified that generic terms combined with a domain name (e.g., "Booking.com") could be trademarked if they acquire distinctiveness. This principle could extend to AI-generated branding elements, provided they meet distinctiveness requirements.

Copyright and AI

Copyright law faces significant challenges with AI-generated works:

• Authorship: The Copyright Office has consistently maintained that only works created by humans are eligible for copyright protection. This was reinforced in the recent Thaler v. Perlmutter case, where the court ruled that AI-generated works without human authorship are not copyrightable.

• Derivative Works: If AI modifies an existing copyrighted work, determining whether the output is a derivative work or an entirely new creation can be complex. This is particularly relevant for generative AI models like DALL-E or ChatGPT.

• Fair Use: AI systems trained on copyrighted material may raise fair use issues. Courts have yet to fully address whether training AI on copyrighted datasets constitutes fair use, but cases like Authors Guild v. Google (concerning Google Books) provide some guidance.

Practical Implications for AI Innovators

Given these legal standards, companies developing AI technology should take the following steps to ensure strong IP protection:

1. Maintain Records of Human Contributions: Document the inventors’ role in AI-assisted discoveries to avoid issues with inventorship.

2. Draft Claims Strategically: Frame claims to focus on AI applications that provide real-world technological improvements rather than just algorithmic processing.

3. Strengthen Disclosures: Provide detailed explanations of the AI’s structure, training methodology, and technical implementation to satisfy enablement requirements.

4. Consider Trade Secrets as an Alternative: If an AI model is difficult to explain in a patent application, keeping it as a trade secret may be a better approach.

5. Trademark Searches: Use AI tools to conduct thorough trademark searches but ensure human review to avoid errors.

6. Copyright Policies: Establish clear policies on authorship and ownership for AI-generated works, and address potential fair use issues in training datasets.

Conclusion

The USPTO’s new AI guidance brings much-needed clarity to AI-related patent applications. It reinforces the requirement of human inventorship, tightens the standards for patent eligibility, and underscores the importance of clear disclosures. Meanwhile, trademarks and copyrights face their own AI-related challenges, from authorship to infringement risks.

For businesses investing in AI innovation, staying ahead of these regulatory developments is crucial. A well-structured IP strategy—balancing patents, trademarks, copyrights, and trade secrets—will be key to leveraging AI technology in the years to come.

If an AI model is difficult to explain in a patent application, keeping it as a trade secret may be a better approach.