In this exceptional situation, we are sharing an ai fray article with you and the other subscribers to ip fray's emails because this is of interest to many patent practitioners.

https://aifray.com/ai-patent-eligibility-spotting-hallucinations-over-value-adds/

AI patent-eligibility: spotting hallucinations over value-adds

Olivia Sophie Rafferty

June 20, 2025

Context: Artificial Intelligence systems are increasingly present in our everyday lives, and have become a popular tool in research and development, too. AI is now being used to accelerate discovery, design software, and even shape new chemical compounds. Now, with the emergence of AI patent drafting tools such as Deep IP, Patent Bots and Solve Intelligence (September 6, 2024 Solve Intelligence blog post), patent professionals are also being able to leverage AI to streamline every step of their work, from checking they comply with specific jurisdictions to full document reviews.

What’s new: When engineers and patent professionals become over-dependent on AI, they “risk more than disappointment”, according to Dr.-Ing. Robert Klinski, a patent attorney and founder of Germany-based firm PATENTSHIP. In a recent LinkedIn post, Dr. Klinski broke down some of the key ways in which AI can mislead innovators, such as through hallucinations (fabricated concepts, fake prior art, or unworkable ideas masked by technical jargon: Dr.-Ing. Robert Klinski LinkedIn post). Some reasons for why AI fails, he writes, include:

• The training data is noisy, biased, or synthetic;
• AI can’t connect abstract ideas or validate feasibility; and 
• An invention needs more than a word prediction model.

Direct impact: The consequences of AI hallucinations are “real”, he warns. In fact, the effect of AI on IP extends beyond the authorship of content produced by AI systems, the data (often subject to copyright) it uses to train such systems, and the use of technology to spread disinformation. Some key consequences, Dr. Klinski writes, include:

• Speculative inventions: chasing speculative or per se “obvious” inventions, or wasting prosecution on invalid or flawed ideas;
• Waste of time: pursuing patents for flawed inventions, discarding valuable inventions due to fake prior art, and sifting through a myriad of AI-generated inventions all burn through hours of valuable time; and
• Loss of competitive edge/profits: poor IP scoping can lead to a missed competitive edge, or erode your ROI in your innovation pipeline.

Wider ramifications: The effects of AI on patent prosecution also came up during a workshop hosted by Patently in London yesterday, which Ben Maling of EIP wrote about on LinkedIn. In a nutshell, he wrote, to reap the benefits of AI safely, attorneys will need to sharpen their reviewing skills; “AI-generated b*llocks may be harder to detect than trainee-generated b*llocks!” (Ben Maling LinkedIn post). Enforceability risks are also a major concern in the community, which Dr. Klinski has labelled as the “enforceability trap”. To ensure that patent professionals build commercially effective portfolios, they must align with European Patent Office (EPO) standards and understand the operational division of AI processes, he wrote earlier this week (PATENTSHIP Patent Attorneys LinkedIn post).

According to Dr. Klinski, the core of the issue lies in the way AI-generated inventions are not always real. “Hallucinations,” he says “, aren’t just bugs, they’re expensive pitfalls.”

This is how IBM defines “hallucinations”:

A phenomenon where, in a large language model (LLM) often a generative AI chatbot or computer vision tool, perceives patterns or objects that are nonexistent or imperceptible to human observers, creating outputs that are nonsensical or altogether inaccurate.

In the context of patent prosecution, hallucinations mean AI may generate false positives, such as inventing prior art that does not exist or omitting critical known prior art, which skews the scope and prosecution of those patent applications. Often arising from noisy, biased training data, hallucinations mean AI cannot “truly” connect the dots across domains, nor can it judge feasibility hidden behind complex DSP algorithms or hardware constraints.

Dr. Klinski and Mr. Maling are not alone in voicing their concerns about AI and patent prosecution. Blueshift IP’s Robert Plotkin also wrote recently about how today’s AI systems add a deceptive layer of sophistication to invention disclosures, a phenomenon he likes to call “garbage in, polished garbage out” (Robert Plotkin LinkedIn post).

While acknowledging that AI tools can be valuable in the patent process when used appropriately, he warns that AI often takes basic input from inventors and transforms it into text that appears thorough and professional but lacks the substance necessary for effective patent protection.

This leads to three main problems:

1. AI cannot identify and emphasize the truly novel aspects of an invention;
2. AI-generated disclosures are often significantly longer than they need to be, which wastes time and creates inefficiency;
3. The “false confidence effect” of polished AI documents leads patent professionals to think they can make business decisions about patentability or commercialization – but these will often be based on an inadequate understanding of the invention’s unique value.

Mr. Plotkin recommends these three ways to leverage AI more effectively in drafting and prosecuting patents:

1. Reverse the Role of AI: Create your own focused description of the key innovative elements first and then use AI as a checking or editing tool to improve clarity, not to add volume.
2. Focus on Points of Novelty: Before involving AI, identify what you believe makes your invention patentable. What specific elements are novel and non-obvious?
3. Use AI for Specific Tasks: Rather than general disclosure generation, use AI for more focused tasks like:

• Suggesting alternative phrasing for technical concepts;
• Helping explain complex processes more clearly;
• Identifying potential applications you may have overlooked;
• Checking for internal consistency in your descriptions; and
• Having AI ask you clarifying questions about your invention to draw out your expertise.