Patent Considerations for AI-Assisted Inventions

AI cannot be an inventor

The previous section noted that no major vendor indemnifies against patent claims on AI-generated output. Code functionality can infringe patents regardless of how it was written. But a related question arises: when AI assists in developing a potentially patentable invention, who qualifies as the inventor?

Only natural persons humans can be named as inventors on patent applications. AI systems cannot be inventors, joint inventors, or receive any inventorship credit. This principle has been tested globally and rejected in every major jurisdiction.

The Thaler v. Vidal precedent

Stephen Thaler developed an AI system called DABUS (Device for the Autonomous Bootstrapping of Unified Sentience) and filed patent applications in multiple countries listing DABUS as the sole inventor. His applications covered a food container design and an emergency lighting beacon. Thaler claimed DABUS autonomously conceived these inventions without human assistance.

The U.S. Federal Circuit Court of Appeals decided Thaler v. Vidal in August 2022. The court unanimously held that the Patent Act requires an inventor to be a natural person. Statutory terms like "individual" and "whoever" historically refer to humans, not machines. Thaler's argument that allowing AI inventors would encourage innovation was dismissed as "speculative and lacking a basis in the text of the Patent Act."

The U.S. Supreme Court declined to hear the case, leaving the Federal Circuit's ruling as controlling law.

Global consensus

Patent offices worldwide reached the same conclusion:

South Africa initially granted a DABUS patent, but this reflects administrative practice rather than judicial determination. The global judicial consensus is unanimous: only humans can be inventors.

USPTO November 2025 guidance

The U.S. Patent and Trademark Office issued revised inventorship guidance on November 28, 2025. This guidance replaced and rescinded the February 2024 guidance entirely.

AI as a tool

The November 2025 guidance treats AI systems as tools analogous to laboratory equipment, computer software, research databases, or any other instrument assisting the inventive process. No separate standard applies to AI-assisted inventions. The same legal analysis governs all inventorship determinations, regardless of whether AI was involved.

This differs from the February 2024 approach. The earlier guidance applied special scrutiny to AI-assisted inventions and required demonstrating "significant contribution" using the Pannu factors. The current guidance eliminates that requirement for scenarios where a single human inventor uses AI.

The conception standard

Inventorship centers on conception the formation of a "definite and permanent idea of the complete and operative invention." The human inventor must possess a specific, settled idea with a particular solution. A general goal or research plan is insufficient. The invention must be "so clearly defined in the inventor's mind that only ordinary skill would be necessary to reduce the invention to practice, without extensive research or experimentation."

AI systems cannot conceive inventions because they are not persons. When humans use AI tools, the human remains the inventor if they conceived the invention, regardless of the AI's contribution to implementation or exploration.

Joint inventorship unchanged

When multiple humans collaborate on an AI-assisted invention, traditional joint inventorship analysis still applies. Each joint inventor must make a significant contribution to conception of at least one claim. AI tools factor into this analysis only as instruments that humans used they do not themselves qualify for any inventorship analysis.

Documentation requirements

Incorrect inventorship can render patents unenforceable or invalid. While the USPTO's November 2025 guidance lowered barriers compared to the February 2024 approach, the underlying legal risks remain. Courts have not yet ruled on how much human contribution is required when using AI. Contemporaneous documentation provides the evidence needed to defend inventorship if challenged.

What to document

Maintain records that establish the human inventor's conception:

Problem framing and design choices. Document the inventor's thinking before engaging AI tools. What problem was being solved? What approaches were considered? What constraints shaped the design? These decisions often occur before AI is involved and establish human direction of the inventive process.

Prompts and interactions. Record what inputs were provided to AI tools. Document configurations, training data choices, and parameter selections. These records show human direction rather than passive acceptance of AI output.

Selection and modification. When AI generates multiple possibilities, document why specific outputs were selected. Record modifications made to AI suggestions. The act of recognizing which AI output constitutes an invention and improving upon it demonstrates human conception.

Claim development. Track how patent claims evolved. Document which claim elements the human conceived versus which elements emerged from AI exploration. The claims define the invention; demonstrating human contribution to each claim element strengthens inventorship.

Inventor notebooks

Traditional inventor notebooks remain valuable for AI-assisted work. Describe the complete and operative invention as conceived by the human inventor. Record this description contemporaneously, not reconstructed after filing. Date entries and maintain continuity.

Position AI outputs like laboratory results. A chemist documents their hypothesis, runs experiments, then interprets results. Similarly, an inventor using AI documents their conception, uses AI tools, then interprets and refines outputs. The AI is an instrument; the inventor interprets what the instrument produces.

Invention disclosure forms

Update disclosure forms to capture AI involvement:

• Require identification of all natural-person contributors only
• Include fields for documenting AI tools used
• Require description of human contributions to each inventive element
• Document modifications made to AI outputs
• Capture the inventor's problem framing and design decisions

Resolve inventorship questions during disclosure, before drafting begins. Correcting inventorship after filing creates complications best avoided.

Enterprise audit trail practices

Beyond individual documentation, enterprises need systematic approaches to tracking AI involvement in R&D.

Version control for AI-generated materials

Treat AI outputs like any other development artifact. Store prompts, AI responses, and human modifications in version control. Maintain audit trails linking development history to patent applications. If inventorship is challenged, these records demonstrate the human creative process.

Inventor interviews

During patent preparation, conduct structured interviews to document contributions. Ask each inventor to identify their specific contributions to claim elements. Record these interviews. For AI-assisted inventions, explicitly document how AI tools were used and what the human invented beyond the AI's output.

Internal audits

Audit inventorship determinations periodically, particularly for AI-heavy development projects. Review whether disclosure forms capture AI involvement accurately. Verify that named inventors can articulate their conception independent of AI assistance. Identify patterns where inventorship documentation may be weak.

Foreign filing considerations

Priority claims to foreign applications must include at least one overlapping natural person as inventor. If a foreign application listed AI as an inventor (as early DABUS filings attempted), U.S. applications cannot claim priority from it. Ensure global R&D teams understand U.S. inventorship requirements before filing anywhere.

Practical implications for ASD

Developers using AI coding agents will occasionally create patentable inventions. A novel algorithm, an innovative data structure, an inventive approach to a technical problem these can arise during agent-assisted development just as they arise during traditional coding.

Recognition matters

The human developer must recognize when something patentable emerges. Agents do not flag potential inventions. If a developer directs an agent to solve a problem and the solution involves genuinely novel technical innovation, the developer should document their conception of that innovation not merely accept the output.

The contribution question

Prompting an agent with "solve this problem" and accepting whatever emerges may not establish adequate human conception. Courts have not yet defined the threshold, but prudent practice involves documented human contribution:

• What problem definition shaped the approach?
• What constraints did the human specify?
• What selection criteria drove choosing this solution over alternatives?
• What modifications improved the AI's initial output?

These contributions establish the human as inventor, not merely the operator of an AI system.

Patent risk parity

AI-generated code can infringe existing patents just as human-written code can. The generation method does not affect infringement analysis. No vendor indemnifies against this risk. Enterprises face the same patent exposure with AI-generated code as with human-written code, but with less visibility into why a particular implementation was chosen.

Patent clearance processes searching for potentially infringed patents before shipping apply to AI-generated code without modification. AI-generated code may warrant closer scrutiny because developers have less intuition about why the code takes a particular approach.

What happens next

Patent law has not fully adapted to AI-assisted invention. The USPTO's November 2025 guidance represents current policy, but policy can change. Courts have not addressed many practical questions about AI and inventorship. Congressional action could modify statutory requirements.

The legal framework governing AI-assisted inventions will continue evolving. Documentation practices that seem conservative today may become minimum requirements tomorrow. Building robust audit trails now provides protection against future legal developments.

The practical guidance: document human conception thoroughly, treat AI as a tool rather than a co-inventor, and maintain records that can demonstrate inventorship if challenged. Patent protection remains available for AI-assisted inventions. What matters is establishing that a human not the AI conceived the invention.