AI and the R&D process: Why inventors must rethink how they use it before filing a patent application

Artificial intelligence is becoming part of everyday workflows in research and development.  

From drafting technical descriptions to summarising ideas, it can seem like a helpful shortcut in a fast-paced innovation environment.

However, conversations with patent attorneys highlight a growing concern: the way inventors are using AI is starting to create real risks to both the quality of inventions and the ability to protect them.

The issue is not just how AI is used, but when and what is shared.

The illusion of progress: When AI produces more words, not better inventions #

One striking trend is the increasing volume of AI-generated material being shared with patent attorneys. Inventors are arriving with long, polished documents that appear detailed and professional, but often lack substance.

This creates a new challenge:

• inventors may struggle to explain their invention clearly in their own words
• the underlying idea is sometimes not fully developed or even workable
• the volume of content makes it harder (not easier!) to identify what is actually inventive  

In some cases, the reliance on AI is so strong that inventors cannot describe how their solution works, casting doubt on whether they can actually be identified as an inventor in any patent application.

For R&D teams, this is a critical signal: AI can give the impression of progress, but it does not replace genuine technical understanding or experimentation.

Why disclosure risk is the real concern #

Beyond quality, the most serious issue highlighted in the discussion is disclosure.

When inventors input technical details into public AI tools (such as large language models), there is uncertainty about how that information is handled. As a precaution, anything entered into such systems should be treated as no longer confidential.  

Even if the AI tool does not explicitly “repeat” the invention to other users, there is a risk that:

• the information contributes to the system’s training
• elements of the idea could appear in responses to other users’ prompts
• ownership and origin of the idea become highly difficult to trace  

From the perspective of patent attorneys, this creates a potential problem, since if an invention is considered to have been made available to the public, it may no longer be patentable.

A simple principle for inventors: Assume nothing stays secret #

For inventors working in R&D, the safest approach is to not enter confidential invention details into AI tools before a patent application is filed, unless confidentiality is explicitly guaranteed.

This applies especially when:

• describing how a technology works
• generating or refining solution concepts
• preparing invention disclosure forms or technical summaries

Even internal processes can create unintended behaviours. For example, it may be tempting to use AI to produce more polished invention disclosure documents, but this can backfire and result in potential pre-filing disclosure, that renders the invention not novel or leads to a situation where the invention is deemed not sufficient enough to be enabled.

In view of the pitfalls, we are increasingly hearing from client companies explicitly rejecting internal AI-generated submissions, requiring inventors to describe ideas in their own, authentic words and with their voice of project experience. We have also noticed that many companies are now asking inventors to list the AI tools they have used in the implementation in order to track the use and input of AI into any eventual patent portfolio.

Why patent attorneys prefer direct interaction, not AI-generated content #

Another key insight from discussions with patent attorneys is how they work in collaboration with inventors. Rather than relying solely on written documents, attorneys also speak directly with the inventors, inviting them to describe the invention in their own words. These informal discussions enable iterative questioning, helping both parties to identify the aspects of the invention that are truly novel and patentable.  

Material that is solely, or heavily, AI-generated can interfere with this process. It may introduce unnecessary complexity, emphasise the wrong aspects of the invention, or anchor discussions around content that is not technically meaningful. In practice, this can make it harder for attorneys to identify the genuinely novel aspects of the invention, slowing down the patent drafting process and potentially reducing the quality of the resulting patent application.  

What AI means for R&D teams #

AI is not inherently problematic, but it needs to be used in a controlled and informed way. To help minimise risk while benefitting from AI’s capabilities, R&D teams should focus on three priorities:

• Protect confidentiality: Treat any input into AI systems as potentially public. Avoid sharing invention details before filing.
• Maintain ownership of the invention: Ensure inventors can clearly explain their work in their own words. AI should not substitute technical understanding.
• Use AI selectively: AI can support formatting, summarisation, or administrative tasks, but should not drive the inventive process itself.

AI can be a powerful tool, but strong patent protection still depends on capturing the inventor's expertise, protecting confidentiality, and preparing high-quality patent applications. If you would like to discuss how AI can be used safely within your innovation and patenting processes, please contact your usual Murgitroyd attorney or a member of our patent team.