Due diligence on intellectual property (IP) used to follow a fairly predictable script. Buyers, investors, and partners checked ownership records, confirmed the scope of coverage, and ensured the portfolio aligned with the company’s strategy. Those steps are still important. But AI has introduced new legal questions that don’t fit neatly into the old framework. With courts continuing to refine the rules, diligence teams that stick to the traditional checklist risk missing issues that could materially affect a transaction. In fact, Federal courts have spent the past two years applying a new level of scrutiny to AI patents that borders on skepticism. The pattern is consistent too. Claims that describe AI applied to a business problem without describing any real technical advance are being struck down.
This isn’t just a hurdle for young startups, it also impacts legacy enterprises integrating AI into existing products. When patent claims amount to little more than longstanding business processes automated by AI, the risk is identical, and that’s a huge problem for AI and any company that uses it. If a portfolio fails to demonstrate a genuine technical improvement, its valuation plummets.
The Foundational Test
The impact of AI doesn’t make the fundamentals optional. Deals will crumble over foundational gaps long before anyone reaches a legal eligibility argument. It’s why today’s diligence requires more than checking the traditional boxes, and why these three questions now deserve far more attention during IP diligence:
- Does the portfolio match the business? A patent portfolio should reflect what the company actually makes, where it’s investing, and how it expects to grow. If the patents don’t line up with the business, buyers will want to understand why. That review should also include freedom to operate. Companies that have assessed the surrounding patent landscape and identified potential third-party rights are in a much stronger position than those encountering those issues for the first time during diligence.
- Is the portfolio in good order? Even technically strong patents lose value if foundational ownership and formalities are not in order. It’s imperative to periodically confirm accurate inventorship or pending and issued patents, ensure assignments and declarations are properly executed and recorded and maintain current maintenance fee payments across relevant jurisdictions. Failure to confirm this can affect valuation and delay transactions, and for no good reason. So, make sure to confirm the above.
- Is valuable IP being protected outside the patent portfolio? Not every valuable asset should, or can, be patented. Proprietary data, know-how, source code, and internal processes often derive their value from remaining confidential, which means confidentiality policies and trade secret practices deserve just as much attention as the patent portfolio. The same is true for open source software. Companies should understand what they’re using, comply with applicable license obligations, and have processes in place to manage those risks. Keep in mind source code review done at deal time can impose huge burdens, delay closing, and reduce value.
Given the deeper diligence question is not whether a company uses AI, virtually every technology company does, companies must (also) address whether its IP portfolio captures genuine technical innovation. This includes how that AI is built, trained, deployed, or improved. And the courts feel this way too.
Why Courts Are Getting Tougher on AI Claims
Judges are now drawing a hardline pushing patent holders for answers. Does a patent improve how the AI works? Or, does it just use existing AI in a new field? Claims in the second category are routinely being struck down. The Federal Circuit recently made this benchmark clear in Recentive Analytics, Inc. v. Fox Corp, which addressed patent claims directed to the use of machine learning to optimize television network schedules and map generation.
The court found the claims patent-ineligible, and held that they were directed to abstract ideas, with no meaningful technical improvement beyond the application of machine learning to those existing problems. It was not that machine learning itself was abstract; it was that the claims failed to recite any specific improvement to machine learning technology.
A Second Warning From the Patent Office
Simply applying AI to a different use case isn’t enough to qualify for protection either. A portfolio packed with “AI applied to X” patents is far less valuable and enforceable than raw patent counts suggest.
The Federal Circuit isn’t the only place this issue has come up. The U.S. Patent and Trademark Office has also rejected AI-related patent applications for many of the same reasons, often before a patent is ever issued.
In re Brian McFadden, an applicant claimed AI-based methods in broad terms without explaining what made the technology itself new or different from conventional computing. The Patent Trial and Appeal Board rejected the application on similar patent-eligibility grounds. Taken together, it’s clear that describing an invention as AI-powered isn’t enough to support a patentable claim. When evaluating an AI patent portfolio, the key question isn’t whether the invention uses AI, it’s what the claims actually protect.
What Defensible AI Patents Look Like
Therefore, diligence on an AI company now also requires a technical conversation and a deeper look. This is why reviewers must evaluate whether patent claims map to tangible architecture rather than generic business outcomes. Though not every AI patent is vulnerable. Diligence also has the ability to identify real strength.
So, to identify true portfolio strength, diligence teams should be sure to identify innovation in:
- Architecture: Structural innovations, such as new focus mechanisms, memory designs, or hybrid model configurations, can support strong claims when described in detail.
- Measurable performance gains: Claims tied to specific engineering choices, such as a preprocessing step that boosts robustness or an inference optimization that lowers latency without hurting accuracy, reflect real technical substance. In another example, reducing hallucinations over conventional models and architecture can demonstrate technical contribution. This is key to demonstrate rather than a repackaged application of existing tools.
- System-level design: Innovation doesn’t have to live inside the model itself. Data pipeline design, hardware-software interfacing, privacy-preserving inference, and security architecture can all support defensible claims outside the abstract idea problem.
What This Means in Practice
Standard diligence checklists are still necessary, and will be for some time but they were built for a period when patent eligibility was rarely in serious doubt. That period has passed. Any patent application should include an invention story that ties a technical contribution to the claimed invention. Inventors can help define and educate on such technical contributions. The claimed invention should be carefully crafted to protect strategic targets for the business. Identifying and validating this balancing act is the goal of the modern diligence review in addition to classic review. Those properties that effectively walk this high-wire prove their worth, while the rest leave questions on enforceability and validity that may sacrifice value.
