[Legal Insight] "My AI Invented This": Navigating Inventorship Risks Post-DABUS (US & EP Perspective)
Introduction: The "ChatGPT" Question
A few years ago, the question "Can an AI be an inventor?" was a philosophical debate for legal scholars. Today, it is a practical question asked by clients in initial disclosure meetings.
With the conclusion of the DABUS saga (Stephen Thaler’s attempt to name an AI as an inventor), the highest courts in the US, UK, and the EPO have spoken with a unified voice: "No. Inventors must be natural persons."
However, as patent attorneys, we know the "No" is the easy part. The real challenge arises when a client asks:
"Okay, I can't name the AI. But the AI generated 90% of the code, and I just tweaked it. Can I still file this as the sole inventor?"
This post goes beyond the DABUS headlines to discuss the practical risks of "AI-Assisted Inventions" and how to counsel clients to ensure their patents withstand validity challenges in the generative AI era.
1. The Judicial Consensus: The Door is Shut for AI Inventors
Before diving into the gray areas, let’s briefly confirm the black-and-white rules established by recent case law.
United States (Thaler v. Vidal): The Federal Circuit affirmed that the Patent Act (35 U.S.C.) defines an "individual" as a natural person. AI cannot be named as an inventor.
Europe (EPO J 8/20): The Board of Appeal confirmed that under the EPC, the designated inventor must be a human being. The designation must contain a family name and full address.
UK (Supreme Court - Thaler): Unanimously dismissed the appeal, ruling that an inventor must be a natural person under the Patents Act 1977.
The Takeaway: Naming an AI on the Application Data Sheet (ADS) or Request for Grant form is a procedural error that will lead to immediate rejection.
2. The New Danger Zone: "AI-Assisted" & Invalidity Risks
The risk has shifted from "Can AI be an inventor?" to "Is the human named truly the inventor?"
The USPTO recently issued crucial "Guidance on AI-Assisted Inventions" (Feb 2024). The core principle is the "Significant Contribution" test (derived from Pannu factors).
The Risk Scenario:
If a client claims an invention where the AI performed the "mental part" (Conception) and the human merely acted as a supervisor or prompted it with a generic request (e.g., "Design a better antenna"), naming the human as the sole inventor could constitute Improper Inventorship.
In the US: Incorrect inventorship can render a patent invalid (35 U.S.C. § 102/115) and potentially unenforceable if deceptive intent is found.
In Europe: While the EPO is less investigative about entitlement during examination, entitlement disputes in national courts could strip the patentee of their rights if the true "deviser" is not the human applicant.
3. Practice Tips: How to Secure "Human Contribution"
When a client arrives with an AI-heavy invention, you must dig for the "Human Spark." Here is how to frame the narrative to satisfy the USPTO and EPO requirements.
Tip 1. Prompt Engineering as "Conception"
A generic prompt ("Make this faster") is not conception. However, a highly technical, iterative prompt can be.
Argument: The human inventor constructed a specific prompt containing technical constraints, boundary conditions, and architectural parameters. The AI acted merely as a "sophisticated calculator" executing the human's specific design plan.
Action: Ask the client for the prompt logs. If the prompts demonstrate specific technical direction, the human is the inventor.
Tip 2. Selection and Verification (The "Director" Theory)
The USPTO guidance suggests that merely "recognizing" an AI output is good is not enough. However, if the human "takes a specific output and significantly modifies it" or "runs experiments to verify a specific property," that creates a human contribution.
Drafting Strategy: In the specification, emphasize the human's criteria for selecting the output. "The inventor identified that Candidate Model B failed to meet the latency requirement, and thus modified the architecture by..."
Tip 3. Focus Claims on the "Human-Controlled" Aspects
Ensure the independent claims recite features that directly map to the human's contribution.
If the AI generated the code, but the human designed the system architecture or the data processing logic, draft method claims focusing on the logic/architecture rather than the specific code implementation.
4. Future Outlook: The "German Compromise"?
While the US and UK took a hard line, the German Federal Patent Court offered an interesting middle ground:
The inventor must be human.
BUT, the applicant is allowed to state in the description that "Artificial Intelligence [Name] was involved in generating the invention."
This approach preserves the legal requirement of a natural person while satisfying the applicant's desire (or moral obligation) to disclose the AI's role. We may see more jurisdictions adopting this "transparency" approach in the description, even if the ADS remains human-only.
Conclusion: You Are the "Inventorship Auditor"
In the age of Generative AI, the Patent Attorney's role expands to become an "Inventorship Auditor."
We must educate clients that AI is a powerful tool, but patent law rewards Mental Conception, not just computational output.
When a client asks, "Can I patent this AI-generated idea?", your answer should be:
"Yes, provided we can prove that YOU were the architect and the AI was merely the builder. Let's look at your prompts and your verification process."
📝 Practitioner’s Checklist
[ ] ADS Check: Ensure no AI is listed in the inventor field (avoid "DABUS," "ChatGPT," etc.).
[ ] Inquiry: Did the client use GenAI? If so, for which parts? (Code, Molecule Candidate, Schema).
[ ] Assessment: Does the human contribution meet the "Significant Contribution" standard (USPTO)?
[ ] Evidence: Advise clients to retain Prompt Logs and Iteration History as evidence of conception (priority records).
[ ] Disclosure: In the US, consider the Duty of Disclosure regarding the extent of AI use if it impacts inventorship analysis.
