Author: Patent Consultant Maria Laini
In an interview released a few months ago (1), OpenAI CEO Sam Altman raised concerns about the lack of a confidentiality legal framework for ChatGPT and other open AI platforms. Altman has confirmed that, differently from your conversations with lawyers, doctors and therapists, there is no legal privilege to protect you when feeding AI tools with your personal information.
This is no surprise. Why would an open, free chatbot be regulated by the same duty of care as a chartered professional?
This needs to be a warning to clients and attorneys alike, as we need to be very mindful of how we use AI tools in our work. A conversation with ChatGPT is not going to be “behind closed doors”, even if nobody is in the room with you when you copy and paste that draft that you desperately want to translate into the chat. Importantly, this also extends to any other web-services such as “Google Translate”. AI does not need to come into the picture, to ring some alarm bells regarding how your data will be used.
This means that any document that is not already public, and anything that should be kept confidential, should not end up in any public and unregulated web-service, AI or not.
What happens if you upload a patent draft on ChatGPT?
You may now wonder, what is the actual consequence of uploading on ChatGPT (or on similar services) that long and highly confidential patent draft of yours.
Unfortunately, there is a very clear legal consequence, which is that your patent draft is now prior art, and therefore, it is not technically patentable anymore.
In patent law, prior art is defined as everything that was made available to the general public at the time you filed your patent application. This does not only encompass articles, patents and patent applications that you can find on official registers and journals, but everything that has been made available to the public, anywhere in the world and in (almost) any circumstance (2).
Crucially, something can be prior art even if nobody has actually accessed that material, at any point in time. What matters is that it was accessible, rather than accessed.
For example, a paper copy of a patent application left on a train would be prior art, so as a book that is only stored in one library in a relatively remote location and in an obscure language. Similarly, conversations made in the open, not protected by confidentiality agreements or any form of legal privilege are also considered prior art. Imagine being in a coffee shop, and talking with your friend about your invention – everything said in that conversation is now prior art. Only disclosures to someone bound by an NDA or any other type of confidentiality duty would be an exception to this.
This means that all conversations with an open chatbot or web-service (AI or not) that does not have a regulated and restricted data usage policy should be treated like public conversations.
If you are a patent applicant, you may wonder if this would, in practice, mean that your patent application will be refused by an examiner.
The truth is that, unless you inform them, it seems unlikely an examiner or a patent attorney will find out that your data has been leaked, and it seems even more unlikely that an examiner will use this against you during examination. However, we must be aware that this is pretty much unchartered territory, at the moment. On the other hand, I can see how a competitor who knows of such data leak may attempt to use this information to revoke the patent, or to oppose it.
Regardless of how small the actual risk of any practical consequence is, a patent attorney will always have the duty to prevent such data leaks, so that your invention does not become prior art as a result of them. It is an important part of our duty of care to ensure that data are always kept confidential, to prevent any fatal outcome for your applications.
Can Patent Consultants use AI tools or is it too risky?
Luckily, this does not mean that we cannot use AI tools to facilitate our job, because there are many AI patent tools out there which have much more restricted and regulated data usage policies. Indeed, when using AI for our day to day work, we have a duty to check the data policy of the AI tool we are using. Fundamentally, we must ensure that:
- the data that we are inputting is not retained
- and/or reused by the LLM used to train the model
- the work we do for different clients is “sealed” and that data is not transferred between, for example, different chats used to draft different patent applications, inside the AI tool.
These are all questions that must be clarified with the AI tool provider, before use.
Conclusion
When using AI tools, it is paramount to use extreme caution in order to protect data confidentiality. One sensible way to do this is to always read the data usage policy of any provider that we intend to use. Finally, since patent attorneys are bound by strict codes of conducts, clients should expect us to use AI responsibly in our day to day job.
(1) https://www.youtube.com/watch?v=aYn8VKW6vXA
(2) The only exception is made for information that was revealed as a result of a breach of confidentiality, although the conditions around this depend on the jurisdiction.
