AI Can

---
title: 'AI Can't Be a Patent Inventor: Japan's Court Got It Right'
date: '2026-07-03'
description: 'Japan ruled AI can't be a patent inventor. After 15+ years building real products, I think they got it exactly right — and here's why.'
author: 'Matthew J. Whitney'
tags: ['artificial intelligence', 'machine learning', 'ai integration', 'llm']
category: 'ai_ml'
published: true
---

The question of who qualifies as an **AI patent inventor** came up in a client engagement about two years ago in a way I'll never forget. We'd built a generative AI pipeline that produced a genuinely novel method for compressing and routing real-time telemetry data. The system — a combination of a fine-tuned LLM and a custom inference layer — had surfaced an approach none of us had explicitly designed. The client's legal team was ecstatic. Their first instinct was to file a patent. Their second instinct, almost immediately after, was to ask: "So... who invented this?"

The room got quiet in a way that rooms only get quiet when nobody actually knows the answer but everyone has a strong opinion about it. One engineer half-jokingly said we should list the model. The IP attorney looked like she'd swallowed something unpleasant. I said what I genuinely believed then and still believe now: the humans in the room who defined the problem, chose the architecture, curated the training data, and interpreted the output — those are your inventors. The model is a very sophisticated tool. A remarkable one. But a tool.

We filed the patent with human inventors listed. Nobody pushed back. But the question never fully went away for me, and apparently it hasn't gone away for the courts either.

---

## Japan's Supreme Court Just Settled It — For Now

Japan's Supreme Court [ruled this week](https://www.japantimes.co.jp/) that an artificial intelligence system cannot be listed as a patent inventor under Japanese law. The ruling aligns Japan with decisions already reached in the United States, the United Kingdom, and the European Patent Office — all of which have rejected attempts to name AI systems as inventors on patent applications. The most famous of these attempts was the DABUS case, in which AI researcher Stephen Thaler spent years litigating across multiple jurisdictions to have his AI system recognized as the inventor of two patents. He lost everywhere.

Japan's ruling isn't surprising from a legal standpoint. But the timing matters. We're in the middle of the most aggressive wave of AI capability claims in the history of the technology industry, and the pressure to extend legal personhood — or at least legal standing — to AI systems is only going to intensify. This ruling is a clear stake in the ground, and I think it's the right one.

---

## The Hype Machine Wants Credit Without Consequences

Here's what's actually driving the push to name AI systems as patent inventors, and it's not philosophical idealism about machine consciousness. It's economics.

If an AI system can be named as an inventor, then the organization that owns and operates that AI system can potentially claim inventorship over an enormous volume of generated intellectual property — without having to demonstrate that any specific human being exercised the creative judgment that patent law has always required. It's a mechanism for laundering automation into ownership at industrial scale.

The AI hype machine has spent the last three years telling anyone who will listen that these systems are not just tools but collaborators, co-creators, even partners. And there's a kernel of truth in that framing — anyone who has worked seriously with modern LLMs knows they can surface ideas and connections that feel genuinely surprising. But "surprising output from a probabilistic system" and "inventive contribution" are not the same thing, and conflating them serves the interests of AI vendors and IP maximalists far more than it serves the interests of actual innovation.

The [Reddit programming community has been grappling with a related concept](https://www.reddit.com/r/programming/comments/1um6xak/we_need_an_accounting_system_for_cognitive_debt/) this week — the idea of "cognitive debt," the invisible accumulation of decisions and dependencies that teams take on when they let automated systems make choices they don't fully understand. It's a useful frame. When you let an AI system be the inventor of record, you're not just making a legal statement — you're offloading accountability into a black box, and the debt that creates doesn't disappear. It compounds.

---

## What "Inventorship" Actually Requires

Patent law in most jurisdictions has a specific and intentional definition of inventorship. It's not about who typed the most characters or generated the most output. It's about who conceived of the claimed invention — who had the mental act of creating or discovering the idea that the patent protects.

This is not a technicality. It's load-bearing.

The conception requirement exists because patents are a social contract. Society grants a temporary monopoly in exchange for public disclosure of how something works. The inventor is the party who made that creative leap, who is accountable for the disclosure being accurate and complete, and who can be held legally responsible if it isn't. An AI system cannot be deposed. It cannot be sanctioned for inequitable conduct. It cannot be held in contempt. It has no legal existence that can bear consequences.

This is why the [United States Patent and Trademark Office](https://www.uspto.gov/initiatives/artificial-intelligence/ai-and-inventorship) has been explicit: AI systems cannot be named as inventors, full stop. The human beings who direct, configure, and apply AI tools in the inventive process — those people can be inventors. The tool itself cannot.

---

## The Accountability Gap Is Real, and It's Getting Wider

I've shipped AI-powered products at scale. Platforms supporting over a million users, systems handling real money, infrastructure where a bad model output has immediate and measurable consequences. And I can tell you from hard experience: the moment something goes wrong, you need a human being who owns it. Not philosophically. Legally, operationally, and practically.

We saw a version of this play out in real time this week when [Reuters reported that Alibaba is moving to ban Claude Code in its workplace](https://www.reuters.com/world/china/alibaba-ban-claude-code-workplace-over-alleged-backdoor-risks-source-says-2026-07-03/) over alleged backdoor risks. Whatever the ultimate truth of that allegation, the situation illustrates something important: when an AI integration creates a security or trust problem, organizations immediately look for a human chain of accountability. Who approved this tool? Who deployed it? Who was responsible for vetting it? The AI system itself is not a party to that conversation.

Now extend that logic to patents. If an AI system is listed as an inventor and the patent turns out to be based on faulty or fraudulent disclosure, who answers for it? If the "invention" infringes on prior art that a human engineer would have been expected to know about, who bears responsibility? These are not hypothetical edge cases — they are the everyday machinery of intellectual property law, and they require human beings who can be held accountable.

The push to make AI a recognized **AI patent inventor** doesn't solve this problem. It creates a legal void where accountability used to live.

---

## The Counterarguments Deserve a Fair Hearing

I want to be honest about the strongest version of the opposing view, because dismissing it doesn't help anyone think clearly.

The most serious argument for AI inventorship isn't that AI systems deserve rights. It's that the current framework creates a perverse incentive: if humans must be named as inventors, then organizations will simply name whoever was in the room when the AI generated the idea, regardless of whether that person contributed anything meaningful to the creative act. The human inventor of record becomes a legal fiction, a warm body attached to a machine's output. Arguably, that's less honest than just acknowledging what actually happened.

This is a real tension. I've seen it. Engineers listed on patents for work they couldn't fully explain, because the system generated it and someone had to sign. It's uncomfortable.

But the solution to that problem is not to eliminate the human accountability requirement. It's to be more rigorous about what constitutes genuine human inventive contribution in AI-assisted workflows. The human who defines the problem space, selects and configures the AI architecture, curates training data, evaluates and refines outputs, and makes the judgment call that this particular output represents a patentable invention — that person has exercised inventive judgment. They're a legitimate inventor. The human who just ran a prompt and copy-pasted the output probably isn't.

Patent law already has tools for this. Obviousness standards, prior art requirements, the written description requirement — these can all be applied with more rigor as AI-assisted invention becomes the norm. The answer is better human judgment, not the elimination of human accountability.

---

## What This Means for Teams Building with AI

If you're building products that incorporate AI — and if you're reading this, you probably are — the Japan ruling and its global equivalents have practical implications you should be thinking about now.

**Document your human inventive contributions.** If your team uses AI tools in your development and innovation process, keep records of the decisions that humans made: the problem framing, the architectural choices, the evaluation criteria, the judgment calls about which outputs were worth pursuing. This documentation is what establishes genuine inventorship if you ever need to defend a patent.

**Don't let "the model figured it out" become a habit of thought.** The [understanding bottleneck is real](https://www.geoffreylitt.com/2026/07/02/understanding-is-the-new-bottleneck.html) — as AI systems generate more of the surface area of our technical work, the humans involved understand less of what's actually happening. That's a problem for patents, but it's also a problem for security, reliability, and maintainability. Build practices that keep humans genuinely in the loop, not just nominally so.

**Get ahead of your IP policy.** If your organization doesn't have a clear policy on how AI-assisted invention is handled — who gets listed as an inventor, what documentation is required, how outputs are evaluated — you're accumulating risk. This is especially true as the legal landscape continues to evolve and different jurisdictions make different calls.

---

## The Correct Call

Japan's Supreme Court got this right. Not because AI systems aren't capable of producing remarkable and genuinely novel outputs — they clearly are. Not because the legal framework for inventorship is perfect and shouldn't evolve — it clearly should. But because accountability, ownership, and legal liability require a human being who can bear the weight of those things. That's not a limitation of AI. It's a feature of civilization.

The **AI patent inventor** question will keep coming back. The technology is only going to get more capable, the outputs more surprising, the human contributions harder to isolate and articulate. Every major jurisdiction is going to have to keep working through this as the technology evolves, and the [USPTO's current guidance](https://www.uspto.gov/initiatives/artificial-intelligence/ai-and-inventorship) will likely need to be revisited as courts encounter new factual patterns.

But the foundational principle — that inventorship requires a human being who can be accountable for the creative act — is not one we should trade away for the convenience of industrial-scale IP generation. The hype machine wants credit without consequences. That's not how engineering works. It's not how law works. And it's not how we build technology that societies can actually trust.

The humans in the room are still the inventors. Let's keep it that way.