AI Patent Infringement Detection

Yes, it works!

Introduction

I’m excited to share the results of our testing of AI tools for patent mining, especially given my long skepticism of software in this space. Even the sophisticated in-house platform we used at Technology, Patents & Licensing (now TechPats) was essentially an organizational tool—useful for rating patents and managing deal flow to acquire assets, but not for true analysis.

With the arrival of real AI, that has changed. These tools are here, and bottom line—they work.

Before diving into results, a quick refresher for readers less familiar with patents. A patent grants the right to exclude others from using the claimed invention. The claim defines, with specificity, what the invention covers. To enforce those rights, the patent holder must show the patent claim “reads on” the accused product. This typically requires examining product literature or reverse-engineering reports and mapping the claim language to product features.

Traditionally, claim-to-product mapping has been handled by specialists with both technical and legal expertise. It is labor-intensive and costly, and even once a mapping is prepared it must be reviewed by a qualified patent attorney to determine whether it truly supports an infringement case. Poor-quality mappings waste attorney time and drive up costs, so analysts must exercise significant judgment in developing them.

Can a machine now do this? Here’s what we found.

The Players

A quick disclaimer: we tested a limited set of vendors based on tool availability. we’ll test others as they come to light and once we have access to their platforms.

Broadly, the players fall into two groups. The “old salts” are long-time IP professionals who understand the realities of patent monetization and have been providing support or consulting services to the industry for some time:

• Limestone (TechsonIP), led by Tom Hochstatter, Luke Barbin, and Chris Mulkey.

• One Transform, founded by Gene Farrelly and Dr. Pouya Taaghol.

• XL Scout, founded by Komal Sharma Talwar.

• PatDel Analytics (ProMax Legal), headed by Aseem Chadha.

On the other side are the newer entrants—the “tech bros” with fresh funding and big ambitions to define the AI patent tool standard:

• Patlytics, founded in 2024 by Paul Lee (CEO) and Arthur Jen (CTO), has raised $21M to date, including a $14M Series A led by Next47 in February 2025. Other backers include Gradient Ventures (Google), 8VC, and Myriad Venture Partners. Patlytics has also assembled a deep bench of seasoned IP professionals, ensuring the product is shaped by real user needs.

• Garden, founded by Adi Sidapara and Justin Mack, has raised $6.8M. While not part of our benchmark per se, they represent another entrant building focused AI tools for patent work.

The Test

We evaluated the tools against a portfolio we know well, selecting a representative patent for analysis. Each system was tasked with identifying potential infringers. In some cases, we ran the tests directly; in others, the vendors executed them. In every case, the output was a results table listing candidate companies, products, and the vendor’s assigned strength of the read. Reports showing the mapping and specific evidence found for a particular company/product can then be generated.

The Results

The table below shows anonymized companies and products, ranked by total score (sum of vendor strength ratings for each claim–product read). Count indicates how many tools showed a possible read.

Company A’s Product A1 leads with a total score of 3.30 and agreement from four of five tools, making it the strongest candidate for further analysis. Other high scorers—B1, C1, and D1—cluster just behind (2.50–2.56) with support from three tools each. Company C’s Product C2 (2.32) also stands out. Collectively, products from Companies A–E dominate the top of the table. Notably, XLScout did not identify Companies A, B, or D at all, yet their products surfaced strongly in other tools.

No single product was flagged by all five tools. However, convergence on Products A1, B1, C1, D1, and C2 points to these as the most productive leads. Beyond the top five rows, scores drop below 2.0 and cross-tool overlap is limited. Products from Company E also appear consistently in rows 6–9 with support from multiple tools. Accordingly, the next step is manual review of products from Companies A–E.

The vendor scoring patterns show calibration differences. Patlytics produced the widest spread (0.50–0.80), acting as the most conservative filter. Limestone, One Transform, XLScout, and Pro Max tended to issue higher or more clustered scores, raising the possibility of inflated or compressed ratings.

Volume is another factor. One Transform generated 344 company-product combinations, many unique to that tool. Interestingly, some of its low-ranked entries were also flagged by Patlytics at modest scores, suggesting weak but non-random overlap. By contrast, Limestone surfaced several entries beyond the top four that no other tool identified. This reinforces the view that the most actionable leads are concentrated in the first five to six rows.

Usability varied. Some tools generated hundreds of pages of raw mappings that required significant effort to interpret. We found the most efficient approach was to apply AI ourselves to extract specifics about the product features identified in those reports.

In short, AI tools clearly accelerate the front end of patent mining, compressing what would take analysts months into hours. But validation remains essential. The next step is to analyze products from Companies A–E to confirm whether they incorporate the claimed technology. That comparison—human verification of AI-identified leads—will reveal both the strengths and blind spots of each tool. We’ll also compare a traditional human powered search and map strategy to the AI tools. Stay tuned.

Good Vibes & Glitches

The advent of AI in this space has created real energy in the industry. Every vendor we worked with was eager to run trials and provide usable results. Since vendors don’t have access to the underlying LLMs for free, I negotiated pricing and paid for each test performed. That said, every system had glitches—it was not uncommon for a platform to fail mid-demo. This seems less about vendor competence and more about the rapidly evolving LLMs and the breakneck pace of new code releases.

Pricing and Availability

Vendors are approaching the market with different strategies. Some offer infringement detection as a stand-alone service or bundled into consulting. Others, like Patlytics, position it as part of a broader patent AI suite. A few even aim to partner with patent holders or integrate the tool into acquisition funds.

Pricing is generally low enough to justify running a small patent set through these tools, though the adage “you get what you pay for” applies. Patlytics—backed by substantial funding and investor expectations—offers the most polished interface and broadest feature set, positioning itself as the leading AI platform for patent work. A seat is not inexpensive, but for companies managing multiple portfolios, the investment may be worthwhile. Consulting groups such as Pro Max provide a cost-effective alternative for quick assessments. Overall, vendors are motivated to compete and are willing to work with customers to get deals done. Go get yourself a deal and try out the tools for yourself.

Conclusions

The benchmark was successful, but results varied enough across tools that cross-comparison was essential. Our test case involved products with abundant public literature, which made analysis easier. Whether these tools can handle more complex evidence—such as reverse-engineering reports with SEM images or ion-beam semiconductor analysis—remains an open question. Their effectiveness in life sciences and pharmaceuticals, where evidence is harder to obtain, also needs to be tested.

In all cases, substantial analyst work will still be required to produce mappings suitable for attorney review. That said, the AI identification of potential infringement and supporting evidence was an overwhelming success. AI tools will undoubtedly shape the future of patent mining and monetization. The real question is how far they can take us—and how much hidden value they will uncover in existing portfolios.

AI tools will undoubtedly shape the future of patent mining and monetization. The real question is how far they can take us—and how much hidden value they will uncover in existing portfolios.