By: Stephen J. Huggins, Partner

Like most attorneys, patent attorneys are struggling with whether and to what extent “AI” tools (e.g., generative tools relying on large language models (LLMs)) can help their clients realize cost savings. Particularly for clients with large patent portfolios, application preparation costs are of central budgetary importance. But there may be an underappreciated legal poison pill that will prevent AI tools from delivering desired savings when it comes to preparation costs.

Attorneys (and agents) investigating AI patent drafting tools often evaluate them based on ethical considerations (e.g., whether the tool protects client information, learns from or otherwise uses client information in impermissible ways, and the like), on performance (e.g., whether the tool generates quality work product without significant/frequent hallucinations), and on cost. 

When it comes to performance, quality varies according to the task the AI tool is being asked to perform, and the array of potential AI tool-generated content is broad. One of the most salient aspirations for a viable task flow is that an AI tool may be trained/fine-tuned on relevant documents (e.g., a client’s own patent portfolio), prompted with a claim set (along with other information), and relied on to generate a quality detailed description based on the training and prompting. Such a flow might significantly reduce costs, and might otherwise appear desirable (assuming the tool meets the more general evaluation standards alluded to above).

But these evaluations and aspirations seem to be missing a key concern, and one which may be practically insurmountable. Using AI tools to draft detailed descriptions – which, again, seems the most likely way to achieve significant cost savings – could make corresponding patents susceptible to new challenges based on incorrect inventorship. To put a finer point on it, replacing a human drafter (i.e., patent attorney or agent) with an AI drafting tool might convert a remote and/or correctable inventorship concern into a fatal, uncorrectable inventorship problem.

Turning now to an abstract example, granted patent claims may include features which arise out of varying degrees of creativity on the part of the drafter, rather than being parroted exactly from the inventor’s own contributions. Such features may have been added from the detailed description into the claims during prosecution. To simplify the example, we assume these additions by the drafter are insufficiently tethered to significant contributions of the inventor, meaning the inventor cannot be considered to have invented the additions.

Historically, failure to name the drafter as another inventor in such situations would not have been considered a serious threat to the patent’s validity or enforceability. For example, the Federal Circuit (responsible for adjudicating patent-related matters) has resisted attempts to challenge patent validity based on supposed inventive contributions by a patent attorney drafter. However, this resistance appears to be grounded on a presupposition that someone – i.e., a natural person who is the inventor or the inventor’s attorney – “invented” the feature at issue. If we instead replace that natural person in the example with an AI tool which legally cannot invent, it seems likely the Federal Circuit will no longer be able to resist such inventorship challenges as it historically has. 

Consequently, using an AI tool to draft patent applications, even if it becomes attractive from a cost savings perspective, could open up a new avenue for challenge on the basis of incorrect inventorship. Moreover, it would seem impractical to address these concerns at the drafting stage by attempting to identify those of the AI tool’s “creative” contributions which go too far beyond the inventor’s own contributions. Such an exercise would be fraught with uncertainty – e.g., relating to application of the Pannu factors – and would likely destroy any desired cost savings from use of the AI tool. 

At least until the language of governing statutes is changed regarding inventorship in natural persons (do not hold your breath), the concerns outlined above could prevent meaningful adoption of AI tools for performing critical patent drafting tasks.

It serves to add one final point, preemptively addressing a seemingly viable rebuttal. One might consider it relatively harmless if an AI tool creates a feature that makes its way into a claim that is deemed invalid/unenforceable for lack of inventorship. After all, the claim might never have been granted, nor the feature described in the application at all, if the AI tool had not produced it in the first place (assuming, for the sake of argument, that the substituted patent drafter would not have added the description in lieu of the tool). But the attempted rebuttal fails to consider the normal patent drafting process. Very often, an inventor gives an initial disclosure, and a patent drafter either prompts the inventor to fill in the missing or omitted feature at issue, or the inventor adds the omitted feature after reviewing the patent drafter’s initial draft. In either case, where the AI tool is the one to add the omitted feature description, it preempts the inventor’s opportunity to contribute the feature for inclusion in the application. 

If you have questions about AI or other patent issues, click to contact Stephen today.