Using AI to Prepare a Patent Filing: What Inventors (and Attorneys) Need to Know

To obtain patent protection in the UK and Europe, an invention has to be kept confidential before filing. As LLMs become ubiquitous, it is increasingly common for inventors to discuss inventions with their AI tools. This article discusses the emerging case law and legal consequences of submitting confidential or privileged information to large language models (LLMs).

In the UK, the Upper Tribunal (Immigration and Asylum Chamber) delivered the first decision by an English court or tribunal that addresses the implications for legal professional privilege when confidential and privileged material is uploaded to open-source AI tools (such as ChatGPT). The decision UK v Secretary of State for the Home Department [2026] UKUT 81 (Hamid) issued a firm rebuke to a UK solicitor who made submissions to the Upper Tribunal that were prepared by a Junior Assistant using ChatGPT that contained fake citations.

 

It also has the following guidance on confidentiality of disclosures to LLMs: “Uploading confidential documents into an open-source AI tool, such as ChatGPT, is to place this information on the internet in the public domain, and thus to breach client confidentiality and waive legal privilege”.

Although these remarks were made in the context of a solicitor putting client letters into ChatGPT, this seems to set a more general precedent (in the UK at least) that a disclosure to an LLM is equivalent to uploading material to a website or similar.

 

In the case of patents, the claims of a patent application have to be novel and inventive to go to grant and any disclosure made before the application was filed can potentially be cited against the patent application by the UK Intellectual Property Office (UKIPO). This includes disclosures made by the patent application holder themselves. We therefore advise clients not to describe an invention in a non-confidential setting to anyone before filing. Traditionally, this has meant no blogging, sales brochures, demos, or YouTube videos of an invention pre-filing, but in light of the above, it would be prudent to assume that disclosures of inventions made to an LLM will also potentially be citable against patent applications.

 

The only exception to this is in the use of “closed source” AI tools. Paragraph [21] of the case cited above says: “Closed source AI tools which do not place information in the public domain, such as Microsoft Copilot, are available for tasks such as summarising without these risks.” The key question is therefore whether an AI tool is scraping data for use in internal training in a way that could lead to regurgitation of an invention disclosure to third parties. The small print should also be checked in case there are clauses that allow for human review of AI prompts and/or data retention which could lead to inadvertent leakage of an invention.

 

Takeaways

In view of the emerging case law, if you want a patent in the UK, don’t input your invention into an LLM without first checking that the particular LLM (and the particular subscription that you have to it) really does keep your invention confidential.