Don’t Let AI Undermine Your Patent: Strategic Dos and Don’ts for Inventors

The use of artificial intelligence (AI) has the potential to be a significant positive force, driving efficiency, innovation, and improved decision-making across many industries including working with intellectual property (IP). However, alongside these benefits come significant risks, and there are simple steps that can be taken to safeguard your IP protection and better support your IP strategy.

Some issues with AI use, such as the problem of false information provided by generative AI, known as hallucinations, is well known. However, there are other reasons to be cautious about how you use AI. These are particularly relevant when it comes to dealing with innovative ideas and legal advice.

Confidentiality

AI tools are  not necessarily confidential. Information provided into Large Language Models (LLMs) and other similar generative AI systems might be stored within the system. This can make the information vulnerable in a number of different ways:

• Just like cloud storage, or other electronic data storage, the data stored and used by AI models can be vulnerable to unauthorised access.
• There are cases of users being able to extract prompts and other information input into AI by other users.
• Data used in training may be reverse engineered from outputs of an AI system.
• The owners/operators of the AI system might be able to infer or extract information from what you enter into an AI system.

All of these points are typical vulnerabilities in computer systems that should be considered as part of your cyber security efforts. However, there can be further, potentially significant, impacts if you are involved in innovating and you want to protect your IP.

Impact on IP

To obtain a patent for an invention, the invention must be novel and inventive. This means the invention cannot have been publicly disclosed in any way prior to filing a patent application.

As mentioned above, data provided to an AI system can be vulnerable, meaning it may be vulnerable to public disclosure . Potentially, information entered into a publicly available AI tool as a prompt may be used to train the AI tool model, and may thus be presented as an output for another user of the AI tool. The simple fact of using an AI model owned and operated by a third party may count as a public disclosure, meaning that the invention is not considered novel and thus cannot be patented.

Common tasks which could cause problems include using AI to expand, develop, or validate your ideas; generating or polishing descriptions or drawings of your ideas; drafting your own patent application; or reviewing patent applications or other documents prepared by your attorney.

Even if patent protection is not something you are interested in, you may want to keep information secret within your company. Trade Secrets, confidential information and know-how can be very valuable IP assets in a business where the value arises in part because no one else knows about them. However, the use of public AI tools with such information might prevent you from doing this, as the information about the idea is no longer under your control.

As a general rule of thumb, unless proprietary information is registered, for example in a filed patent application, it should not be entered into a public AI tool.

Legal privilege and advice

Many people are using AI to help with legal matters. This can be directly seeking legal advice from AI, using AI to help brief your lawyers or distil advice from their lawyers, getting a second opinion on advice provided by their lawyers, or even using AI as your lawyer. However, this can break an important shield around legal advice.

When you are seeking legal advice from a qualified legal professional (such as a patent or trade mark attorney, solicitor or barrister) your discussions are protected by legal privilege. This means that, subject to some very minor exceptions, the information you share with your advisor and the advice they give cannot be used against you in legal proceedings.

AI does not have legal privilege. If you provide information to an AI system it can be (and in some case has been) disclosed to other parties in litigation. This could weaken or fatally damage your case. This can include any questions you ask, any information you give it, or any advice given by the AI.

If you share advice from your lawyer with AI, this may even breach the privilege on that advice, meaning the advice is no longer protected.

Aside from the matter of legal privilege, running legal arguments by an AI tool carries with it the risk of receiving unhelpful advice. Generative AI tools excel at providing very confident and convincing responses even if that response is factually incorrect. This can give users a false sense of security in following AI generated advice which may be wrong, and may give users a trust dilemma if the AI tool generates advice which differs from that provided by a human attorney or other legal adviser.

How to protect yourself

All the above is not to say AI should never be used in a legal or IP context. AI tools can provide excellent support in these fields provided it is used carefully and outputs are critically evaluated.

However,  the same caution should be taken with AI as would be taken in any other situation involving your IP. The maxim of treating everything you put on the internet as a postcard (i.e. once sent, it cannot be un-sent) is just as applicable to everything provided to a public AI tool.

There are steps you can take to mitigate risks in using AI tools:

• Make sure any AI tool you use has proper data security controls and confidentiality in their terms and conditions (many models offer enterprise grade subscriptions that have this built in). That is, a free public AI tool may not be suitable for use with proprietary information, but a local locked-down AI tool may be low risk if the information provided to it is not used to train the model and does not leave the confines of the computing device.
• Use sandboxes or other controlled data environments which reduce the risk of information leaking out from the system.
• Opt out of your information being used for training AI models.
• Avoid certain high risks tasks which may be performed with AI, such as sharing information about a new invention, sharing legally privileged information, and asking for specific legal advice.

However, these steps do not necessarily eliminate the risks and the key point is to always check, and if in doubt, ask. As IP professionals, we would always rather discuss any AI tool use with our clients before the AI tools are used, to help mitigate any risks arising from using the AI tools, and find ways where AI can be used to support IP strategy, not spoil it.

This article was prepared by Partner & Patent Attorneys Oliver Pooley and Janine Swarbrick.