Can AI inventions be patented?

ARTICLE SUMMARY
  • AI technology itself can often be patented if it delivers a novel technical solution.
  • AI-assisted inventions can be patented where there is sufficient human inventive contribution.
  • Fully AI-generated inventions remain legally uncertain, as patent law requires a human inventor.
  • Good record-keeping is increasingly important to demonstrate human contributions and establish inventorship.
authors
Authors
Related Services
Related Industries

Artificial intelligence is reshaping the innovation landscape. AI systems are helping researchers discover new materials, identify drug candidates, optimise engineering designs and solve technical problems at a speed that would have been unimaginable a decade ago. As a result, businesses of all sizes are increasingly asking us the same question … can AI inventions be patented?

The answer depends on what you mean by an "AI invention".

There are really two separate questions. First, can the AI technology itself be protected by patents? Second, can inventions created with the assistance of AI be patented? The legal position is different in each case, and understanding this distinction is increasingly important for businesses investing in AI.

Can the AI itself be patented?

In many circumstances, yes.

There is sometimes a misconception that AI cannot be patented. In reality, patent protection is often available for aspects of the AI technology itself as long as the invention satisfies the normal patentability requirements, in particular the key requirements of novelty and inventive step.

For example, patents may be available for:

  • New machine learning architectures
  • Novel neural network structures
  • Improved training methods
  • Innovative data processing techniques
  • Enhanced inference or optimisation algorithms
  • Technical solutions that improve AI performance, efficiency or reliability
  • Hardware specifically designed to support AI processing

Many of the most valuable AI companies hold extensive patent portfolios covering these types of innovations.

However, obtaining patent protection is not automatic. In Europe particularly, businesses must navigate restrictions relating to computer programs and mathematical methods. An AI-related invention generally needs to provide a technical solution to a technical problem, rather than merely implement an abstract algorithm or provide a merely subjective (rather than objective) improvement.

As a result, patent attorneys will often focus on the technical effect delivered by the AI system. Does it improve image recognition in medical diagnostics? Reduce processor load in autonomous vehicles? Improve industrial quality control? These are the types of technical outcomes where patentability often can be established.

That said, patents are not always the best form of protection for AI technologies.

In many cases, some of the most valuable aspects of an AI system are difficult to reverse engineer and may be better protected as trade secrets. Training methodologies, model weights, datasets, fine-tuning strategies and deployment processes can often derive significant value from remaining confidential rather than being disclosed through a patent application.

For many AI businesses, the most effective strategy combines patents with trade secrets, copyright, database rights and contractual protections. For instance, a patent may protect an invention at the conceptual level (e.g. the AI system architecture), with the specific implementation (weights, datasets, etc.) being maintained as a trade secret.

Can inventions created with AI be patented?

This is where things become more complicated.

The central legal question is whether an invention generated by AI can itself receive patent protection when the inventive contribution comes largely, or entirely, from the AI system.

Recent cases in both Europe and the United States have focused on this issue through the well-known DABUS applications filed by Stephen L. Thaler (a case that led to a UK High Court appeal that Potter Clarkson were directly involved in). In these test cases, patent applications named the AI system DABUS as the inventor rather than a human being. The applications were rejected because patent laws currently require inventors to be natural persons.

The position remains consistent across the major patent jurisdictions. The Federal Circuit in the United States confirmed that inventors must be human beings, while the European Patent Office has repeatedly maintained that a patent application must identify at least one human inventor.

Importantly, however, these decisions do not establish that AI-generated inventions can never be patented. Instead, they establish that the patent system currently requires a human inventor to be identified. The more difficult question is determining how much human contribution is needed (and who has made that contribution) where AI has played a significant role in generating the inventive concept.

AI-assisted inventions are increasingly patentable

Most real-world innovation falls into the category of AI-assisted invention rather than fully autonomous AI invention.

Here, researchers, engineers or product developers use AI tools to accelerate discovery, generate options, test hypotheses or identify solutions. Humans then evaluate those outputs, refine them, select the most promising approaches and transform them into commercial inventions.

In these circumstances, patent protection is often available.

Recent guidance from the United States Patent and Trademark Office makes clear that AI is generally treated as a tool rather than an inventor. The focus remains on whether a human contributed to the conception of the claimed invention.

The practical reality is that most organisations deploying AI in research and development still have substantial human involvement. Scientists and engineers guide the problem-solving process, assess potential outputs, and build workable solutions from AI-generated suggestions.

Provided those human contributions are sufficiently significant, patent protection is likely to remain available.

The growing challenge for patent attorneys

While the legal framework may appear straightforward, proving inventorship is becoming increasingly difficult. Patent attorneys are now being asked a whole range of questions that barely existed a few years ago:

  • Who formulated the prompt?
  • How much freedom was given to the AI?
  • Which outputs were generated?
  • Why was a particular output selected?
  • What modifications were made by the human inventor?
  • At what point did the inventive concept arise?

The answers may become critical if inventorship is challenged during prosecution or litigation.

As AI tools become more sophisticated, maintaining records of the innovation process is becoming increasingly important. Development logs, prompt histories, design records and laboratory notebooks may all play a role in demonstrating the contribution made by human inventors.

Another emerging challenge is inventive step.

Patent offices may increasingly ask whether a solution would have been obvious to a skilled person using widely available AI tools, similarly to how the internet has made it easier to find previously obscure information. The more capable AI becomes, the more difficult it may be to demonstrate that certain advances are genuinely inventive.

So, can AI inventions be patented?

Where the invention is the AI technology itself, the answer is often yes, provided it delivers a patentable technical innovation.

Where AI has assisted humans in developing an invention, patent protection will frequently still be available.

The difficult area remains inventions that are genuinely generated by AI with little or no human inventive contribution. Current law in both Europe and the United States continues to require a human inventor, but the debate around how inventorship should be assessed in an age of increasingly autonomous AI is far from settled. For innovative businesses, that makes early IP strategy, careful record-keeping and specialist patent advice more important than ever.

If you have an AI-related invention and you’d like to explore IP protection, contact us today to arrange a meeting with one of Potter Clarkson’s specialist AI team.

Related News

The Autumn Budget 2026: Why UK innovators must focus on competitive advantage, not just innovation

The Autumn Budget 2026: Why UK innovators must focus on competitive advantage, not just innovation

Can AI inventions be patented?

Can AI inventions be patented?

Taking uterine fibroid research to Parliament

Taking uterine fibroid research to Parliament

Should startups be focusing on investor attractiveness rather than investor readiness?

Should startups be focusing on investor attractiveness rather than investor readiness?

ValueLicensing v Microsoft, GTA 6, and the death of the physical videogame

ValueLicensing v Microsoft, GTA 6, and the death of the physical videogame

Trade mark disputes: choosing your next move

Trade mark disputes: choosing your next move

How can content creators turn social media success into a sustainable business?

How can content creators turn social media success into a sustainable business?

Giulia Avellino on tackling inequalities in uterine fibroid care

Giulia Avellino on tackling inequalities in uterine fibroid care

See All News