Can AI Be an Inventor? Here’s What the Patent Office Thinks

Last updated 2 months ago. Our resources are updated regularly but please keep in mind that links, programs, policies, and contact information do change.

For nearly two centuries, the United States patent system operated on the unspoken assumption that the “inventor” was a human being.

But the arrival of generative AI systems capable of ideating molecules, optimizing mechanical designs, and writing code has challenged this assumption.

U.S. patent policy has moved from skepticism and strict identification of human contribution (the “2024 Regime”) to a new posture of permissive integration where AI is just categorized as a sophisticated instrument (the “2025 Regime”).

The Legal Foundation

To understand the volatile regulatory changes of 2025, you first need to understand the bedrock statutes that constrained the USPTO’s initial responses to AI innovation.

What “Inventor” Means

The central legal hurdle for AI inventorship is found in 35 U.S.C. § 100(f), which defines the term “inventor” as “the individual or, if a joint invention, the individuals collectively who invented or discovered the subject matter of the invention.”

The term “individual” became the focal point of litigation. In the United States, statutory interpretation often relies on the “ordinary meaning” of words. The patent system presupposes a mental act known as “conception”—the formation in the mind of the inventor of a definite and permanent idea of the complete and operative invention. Because “conception” is a mental act, and “individuals” are typically human, the USPTO historically rejected the notion that a machine could conceive.

Thaler v. Vidal: The Case That Set the Rules

The theoretical debate became actual litigation in the case of Thaler v. Vidal. Dr. Stephen Thaler, a pioneer in AI, sought patents for inventions generated by his AI system, DABUS (Device for the Autonomous Bootstrapping of Unified Sentience). Unlike typical AI tools used for optimization, DABUS was claimed to have generated the inventions (a fractal beverage container and a neural flame device) autonomously, without specific prompting.

Thaler listed DABUS as the sole inventor on the application, arguing that accuracy required listing the true creator, even if it was a machine. The USPTO rejected the application. Thaler sued the USPTO Director, leading to a decisive ruling by the U.S. Court of Appeals for the Federal Circuit.

The Federal Circuit’s August 2022 decision established the “black letter law” that remains in force today:

Statutory Construction: The court held that the Patent Act is explicit. The use of “individual” in the statute refers strictly to natural persons. The court cited the Supreme Court’s decision in Mohamad v. Palestinian Authority, which held that “individual” ordinarily means a human being unless Congress clearly indicates otherwise.

Rejection of Policy Arguments: Thaler argued that allowing AI inventors would incentivize innovation and disclosure—the core goals of the patent system. The court rejected this, stating that policy arguments cannot override the plain text of the statute. If the law is to be changed to accommodate AI inventors, it is the prerogative of Congress, not the courts or the USPTO.

The “Thaler Gap”: While Thaler settled the question of sole AI inventorship (an AI cannot be the only inventor), it created ambiguity. It didn’t explicitly resolve the status of AI-assisted inventions where a human and an AI worked together. If an AI contributes 90% of the conception and a human 10%, is the invention patentable? This ambiguity plagued the USPTO throughout 2023 and 2024.

The 2024 USPTO Guidance

Following the Thaler decision, the Biden Administration issued Executive Order 14110 on the “Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence.” This order directed the USPTO to provide guidance on AI inventorship, emphasizing the need to balance innovation with safety and human accountability.

In response, the USPTO issued the “Inventorship Guidance for AI-Assisted Inventions” in February 2024. This guidance represented a “containment strategy,” attempting to preserve the human-centric nature of patent law by rigorously policing the boundary between human and machine contribution.

The Pannu Factors

To determine if a human user of AI qualified as an inventor, the USPTO adopted the test from Pannu v. Iolab Corp. (155 F.3d 1344, Fed. Cir. 1998). Pannu was originally designed to resolve disputes between human joint inventors—specifically, to prevent someone who did very little work from claiming credit alongside the true genius.

The February 2024 guidance repurposed Pannu to evaluate the human-AI relationship. For a patent to be valid, the human named as inventor had to prove they made a “significant contribution” to the conception of the invention. The three Pannu factors required that the inventor:

1. Contributed in some significant manner to the conception or reduction to practice of the invention.
2. Made a contribution to the claimed invention that is not insignificant in quality, when that contribution is measured against the dimension of the full invention.
3. Did more than merely explain to the real inventors well-known concepts and/or the current state of the art.

The Prompting Problem

This framework created a hostile environment for AI-heavy inventions. The guidance explicitly stated that “merely recognizing a problem or having a general goal or research plan to pursue does not rise to the level of conception.”

The most contentious aspect of the 2024 guidance was its treatment of prompting. The USPTO position was that “prompting” an AI—even with a sophisticated prompt—was often just “posing a problem” rather than “conceiving a solution.”

Scenario: A researcher asks an AI, “Design a protein that binds to Receptor X.” The AI outputs a novel protein structure.

2024 Ruling: The researcher posed the problem. The AI conceived the structure. Since the researcher didn’t conceive the specific chemical arrangement, they’re not the inventor. The invention is unpatentable because it lacks a human inventor for the specific claim limitations.

Claim-by-Claim Analysis

The guidance required this analysis for every single claim. If a patent application contained 20 claims, and Claim 5 was generated wholly by AI while Claim 1 was human-generated, Claim 5 was invalid. This forced practitioners to engage in a granular, expensive, and legally risky apportionment of credit for every sentence in a patent application.

The USPTO also clarified that “maintaining intellectual domination” (owning the AI, controlling access to it) was insufficient to confer inventorship. This prevented corporate assignees from claiming inventorship merely by virtue of owning the compute resources.

The Compliance Burden

During 2024, this regime forced applicants to maintain meticulous records separating human thought from machine output. It raised the specter of inequitable conduct allegations; if an applicant failed to disclose that an AI generated a specific sub-routine in a software patent, the entire patent could be rendered unenforceable. This “chilling effect” was criticized by industry groups who argued it penalized the use of advanced tools and put U.S. inventors at a disadvantage.

The November 2025 Reversal

The regulatory landscape shifted seismically in late 2025. Following a change in administration and the issuance of Executive Order 14179 in January 2025, the USPTO engaged in a significant revision of the 2024 framework. The new policy, formalized in the “Revised Inventorship Guidance for AI-Assisted Inventions” published in the Federal Register on November 28, 2025, marks the beginning of the “AI as Tool” era.

Executive Order 14179

The catalyst for the new guidance was EO 14179, titled “Removing Barriers to American Leadership in Artificial Intelligence”. This order explicitly rescinded the Biden-era EO 14110, stating its focus on safety and trust created regulatory requirements that the new administration viewed as barriers to competitiveness.

Key Directives:

Policy of Dominance: The order declares it the policy of the United States to “sustain and enhance America’s global AI dominance.”

Deregulation: It directs federal agencies to review all existing policies and “rescind, replace, or amend” any that act as barriers to AI innovation.

Ideological Neutrality: It mandates the promotion of AI systems free from “engineered social agendas,” signaling a departure from the “algorithmic bias” focus of the previous administration.

The USPTO interpreted the “significant contribution” test of 2024 as precisely such a barrier. By forcing inventors to prove they were “more than just a prompter,” the USPTO was effectively slowing down the patenting of AI-accelerated discoveries.

Rescinding Pannu

The November 2025 guidance formally rescinded the February 2024 guidance “in its entirety.” The USPTO provided a new legal rationale for abandoning the Pannu factors in the context of AI.

The “Sole Inventor” Logic:

The USPTO argued that the Pannu factors are legally applicable only to determine joint inventorship between natural persons.

Since Thaler established that AI cannot be an inventor, an AI system cannot be a “joint inventor.”

Therefore, there is no “joint inventorship” analysis to perform when a single human uses an AI.

Consequently, Pannu is “inapplicable when only one natural person is involved in developing an invention with AI assistance.”

This legal maneuver effectively dissolved the requirement to apportion contribution between human and machine. If the AI is not a legal entity capable of inventing, legally speaking, it contributes zero “inventive” acts. It is merely a passive instrument, regardless of its actual computational complexity.

The “AI as Tool” Doctrine

To fill the void left by Pannu, the USPTO adopted the “AI as Tool” doctrine. The guidance states: “AI systems… are instruments used by human inventors. They are analogous to laboratory equipment, computer software, research databases, or any other tool that assists in the inventive process.”

The Microscope Metaphor:

The USPTO likens Generative AI to a high-powered microscope.

When a biologist uses a microscope to visualize a new bacterium, the microscope “generates” the image. The biologist could not see it with the naked eye. However, the biologist is the sole inventor because they directed the tool and recognized the significance of the image.

Similarly, when an engineer uses AI to generate a new antenna design, the AI “generates” the geometry. But the engineer is the inventor because they directed the AI (prompting) and recognized the utility of the output (selection).

Implications for Prompting:

Under the 2025 guidance, “prompting” is rehabilitated. It’s no longer “mere problem posing”; it’s now viewed as the “operation of the tool.” The specificity of the prompt, combined with the iterative tuning of the AI’s parameters, constitutes the human’s inventive act.

The Presumption: “Don’t Ask, Don’t Tell”

Perhaps the most significant practical change is the presumption of human inventorship. The USPTO now “presumes that the inventors named on the application data sheet or oath/declaration are the actual, human-being inventors.”

This policy, described by legal critics as “Don’t Ask, Don’t Tell”, means that the USPTO will no longer proactively interrogate applicants about the extent of AI usage unless the application file itself contains contradictory evidence (e.g., a specification that explicitly states “The AI system autonomously conceived this embodiment”).

Comparison of USPTO Policy Regimes (2024 vs. 2025)

Subject Matter Eligibility Gets Easier Too

While the inventorship question was being liberalized, the USPTO also reformed Subject Matter Eligibility (Section 101). This is the “gatekeeper” statute that determines if an invention is of the type that can be patented. For years, AI and software inventions faced high rejection rates under the “Abstract Idea” doctrine derived from Alice Corp. v. CLS Bank.

The “Mental Process” Trap

The Alice test prevents the patenting of abstract ideas unless they contain an “inventive concept.” A common category of abstract ideas is the “Mental Process” grouping: concepts that can be performed in the human mind (e.g., basic math, categorization).

Examiners historically rejected AI claims by arguing that neural networks are essentially mathematical operations (matrix multiplication) which, theoretically, a human could perform with a pen and paper, given infinite time.

The August 2025 Memo

In August 2025, Deputy Commissioner for Patents Charles Kim issued a memorandum to examiners. While styled as a “reminder,” the memo effectively raised the bar for Section 101 rejections.

Key Directives:

Practicality Constraint: The memo reminded examiners not to expand the “Mental Process” grouping to include limitations that “cannot practically be performed in the human mind.”

Computational Complexity: Modern AI models involve billions of parameters and operations. The memo clarifies that claims reciting such high-volume computational processes are not mental processes because they’re beyond the practical capacity of the human brain.

The “Tie Goes to the Runner” Rule: The memo instructed that examiners should refrain from issuing a Section 101 rejection unless there is “more than 50% certainty of ineligibility.” This shifted the benefit of the doubt back to the applicant.

July 2024 Guidance Update

The August memo reinforced the July 2024 Guidance Update, which introduced new examples to the Manual of Patent Examining Procedure.

Example 48 (Speech Synthesis): This example clarified that “synthesizing speech waveforms from a cluster of numbers” is eligible subject matter. It’s a technical transformation that the human mind cannot perform.

Example 49 (AI Training): This example likely provided safe harbors for specific methods of training neural networks, distinguishing them from abstract mathematical concepts by tying them to specific improvements in computer functioning.

Impact: The combination of the November 2025 inventorship guidance and the August 2025 eligibility memo has created a “Golden Age” for AI patenting in the U.S. The “who” (inventorship) barrier has been lowered, and the “what” (eligibility) barrier has been narrowed.

Ethical Risks Remain

Despite the “Don’t Ask, Don’t Tell” posture of the USPTO, patent practitioners face significant ethical risks. The Duty of Disclosure (37 C.F.R. § 1.56) requires all parties involved in a patent application to disclose information “material to patentability.”

Materiality Still Matters

Even if the USPTO presumes human inventorship, the Thaler ruling stands: an invention with no human inventor is unpatentable. Therefore, if an applicant knows that an AI generated the invention with zero human direction or review, that fact is material. Hiding it constitutes fraud on the Office.

Scenario: An R&D director sets up an autonomous AI agent to scrape literature and generate novel chemical compounds, then automatically files provisional patents for them. The human “inventor” listed never reviewed the outputs before filing.

Violation: This violates the duty of candor. The human didn’t “conceive” the invention because they never formed the “definite idea” in their mind. The filing is fraudulent.

Consequence: If discovered in litigation (e.g., through discovery of internal emails), the patent would be unenforceable due to inequitable conduct.

The “Black Box” Problem

A subtler risk involves Enablement (35 U.S.C. § 112). A patent must teach a person of ordinary skill how to make and use the invention.

If an AI generates a solution (e.g., a drug structure) but the human inventor cannot explain why it works or how to reliably reproduce the effect without the specific (and perhaps proprietary) AI model, the patent may be invalid for lack of enablement.

The USPTO’s November 2025 guidance does not absolve applicants of the requirement to fully describe the invention. Using AI as a “magic box” to generate claims without understanding the underlying science is a recipe for invalidity.

Practitioner Obligations

The USPTO’s April 2024 guidance on the “Use of AI-Based Tools in Practice” remains relevant. It reminds practitioners that they’re personally responsible for the accuracy of documents submitted to the USPTO.

Hallucination Risk: If an AI drafting tool hallucinates a prior art citation or a technical fact, and the attorney submits it without verification, they’ve violated the duty of reasonable inquiry (37 C.F.R. § 11.18).

Confidentiality: Practitioners must ensure that inputting client data into third-party AI tools (like ChatGPT or specialized legal AI) doesn’t violate client confidentiality or export control laws.

The USPTO Is Using AI Too

The USPTO is not merely regulating AI; it’s adopting it to handle the flood of incoming applications. In October 2025, the Office launched the Automated Search Pilot Program.

Program Details

Start Date: October 20, 2025.

Eligibility: New, original, non-continuing, non-provisional utility applications filed under 35 U.S.C. § 111(a).

Mechanism: Participating applicants pay a fee to have their application subjected to an AI-driven prior art search before it reaches a human examiner.

Goal: To evaluate the impact of sharing automated search results with applicants early, potentially allowing them to amend claims before the first Office Action, thereby speeding up prosecution.

The Rise of AI Prior Art

This pilot highlights a growing challenge: AI-generated Prior Art.

As AI tools become ubiquitous, they’re generating vast amounts of technical disclosure (defensive publications, blogs, code repositories). The USPTO is equipping examiners with AI search tools to find this art.

A key legal battleground will be whether AI-generated prior art is “enabled.” If an AI blog post describes a machine that violates the laws of physics, can it be used to reject a patent application? Current USPTO policy tends to treat published documents as prior art on their face, leaving the applicant to prove they’re non-enabling.

How Different Industries Are Affected

The November 2025 guidance affects different industries in distinct ways.

Pharmaceuticals: AI Drug Discovery

Scenario: A pharma company uses a Generative Adversarial Network to screen 10 million potential molecules for binding affinity to a cancer target. The AI identifies 50 candidates. The human team selects the top 5 for synthesis.

Under 2024 Guidance: The company risked rejection because the AI “conceived” the specific molecular structure. The human only “posed the problem.”

Under 2025 Guidance: The human is the inventor. The human set the parameters (the target), operated the tool (the GAN), and exercised “recognition and appreciation” by selecting the 5 candidates for testing. The AI is simply a high-throughput virtual screening tool.

Software: Coding Assistants

Scenario: A software engineer uses GitHub Copilot to write a complex sorting algorithm. The engineer writes the function header, and Copilot autocompletes 20 lines of optimized code.

Analysis: The engineer is the inventor. They conceived the function of the code and integrated the AI’s output into the larger program. The August 2025 Section 101 memo further protects this invention by recognizing that the complex logic is not a “mental process.”

Design Patents

Scenario: A shoe designer uses Midjourney to generate 500 variations of a sneaker sole. They select one, tweak the tread pattern slightly, and file for a design patent.

Analysis: The 2025 guidance explicitly applies to design patents. The designer’s act of prompting, iterating, and ultimately selecting the specific visual ornamentation constitutes the inventive act. The AI is the “digital pencil.”

The Criticisms

The shift to the “Tool” metaphor has not been universally welcomed. Legal scholars and economists warn of significant downstream consequences.

“Inventorship Theatre”

Professor Dennis Crouch and other IP scholars have criticized the 2025 guidance as “inventorship theatre.” They argue that it maintains the form of human inventorship while abandoning the substance.

By allowing humans to claim credit for outputs they arguably didn’t “conceive” in the traditional sense, the USPTO is engaging in a “legal fiction.” This creates a disconnect between patent law (which rewards human genius) and reality (where machines do the heavy lifting).

Antitrust Risks

The permissive environment creates Antitrust Risks.

Patent Thickets: Large tech companies with massive compute resources can now use AI to generate thousands of patents on minor variations of technology, creating “patent thickets” that block competitors.

Exclusionary Conduct: If a company uses AI to “flood the zone” with patents they barely understand, the FTC or DOJ might view this as an anticompetitive attempt to monopolize a field, distinct from legitimate innovation.

Best Practices for Documentation

In light of the new “Tool” doctrine, the best practices for documenting invention have shifted from “proving contribution” to “proving control.”

Documentation Strategy Post-November 2025

The Prompt Log as Evidence

Patent attorneys now advise treating prompt logs like lab notebooks. In a priority dispute, showing the evolution of the prompt (“Make it lighter” → “Use titanium alloy” → “Optimize for lattice structure”) proves that the human was the driving force behind the design evolution, preventing claims that the invention was “obvious” or purely machine-generated.

What Happens Next

The stability of the November 2025 guidance is not guaranteed. It’s an agency interpretation, not a statute.

Judicial Vulnerability

The Federal Circuit has not yet ruled on whether the “AI as Tool” interpretation is consistent with the Patent Act.

A defendant sued for infringement in 2027 might argue: “This patent is invalid. The USPTO’s 2025 guidance was wrong. The inventor admitted in deposition they just pressed a button. Under Thaler, this is unpatentable.”

If the Federal Circuit agrees, thousands of patents granted under the 2025 regime could be invalidated.

Legislative Intervention

The ultimate resolution lies with Congress. The “AI Patent Act”—hypothetical legislation that would create a sui generis right for AI inventions or explicitly amend § 100(f)—remains a topic of discussion but has not advanced. Until then, the USPTO is bridging the gap with regulation.

The Global Context

The U.S. move to the “Tool” standard aligns it more closely with the European Patent Office, which also requires a human inventor but accepts AI as a tool. However, it diverges from jurisdictions that may remain stricter or become more open to AI-generated rights. The U.S. policy of “Dominance” (EO 14179) suggests that the USPTO will continue to push the envelope of permissibility to ensure American companies can secure IP rights for their AI investments.

As of late 2025, the USPTO considers the use of AI in the invention process not as a disqualifier, but as a standard mode of operation. The burden has shifted from the inventor proving they’re human, to the examiner presuming they are. Yet, the requirement for human conception—the mental spark that ignites the machine—remains the immutable core of American patent law. The AI is the brush, but the human must still be the artist.

Our articles make government information more accessible. Please consult a qualified professional for financial, legal, or health advice specific to your circumstances.