AI Is Coming for Your Life Science Moat: What Strategic Buyers and Private Equity Investors Need to Know About Defensible Enterprise Value

By Daniel J. Holmander and Michel “Mike” Morency, Ph.D. on September 30, 2026

A life science company leaves a trail of public disclosures behind it.  Until recently, piecing that trail together took real time and money.  AI has changed that.  So, the question for a buyer is a different one now: once a competitor has assembled everything the target has made public, what value is left that you can actually own and carry forward?

Key takeaways for deal teams

• Assume a competitor has already used AI to read the target’s patents, papers, posters, trial registrations, job postings, and investor materials together, as one file.
• Patent counts, publication counts, and claims of “proprietary AI” tell you little about defensibility.  What matters is what the company still controls after it has disclosed.
• Durable value now tends to sit in layers: enforceable patent claims, real trade secrets, clean title to data and software, contracts written with AI in mind, and strong regulatory and commercial execution.
• Start this analysis before the term sheet.  Confirmatory diligence is too late.

The diligence question has changed

For most of our careers, the first question in life science diligence was simple: does the target have patents? We now think a better first question is this one.  What could a motivated competitor learn from everything the target has already disclosed, and what could it do with that knowledge using AI?

Patents still matter a great deal, and so do clinical data, software, trade secrets, contracts, brands, and commercial relationships.  The trouble is that none of them can be valued on its own anymore.  Work that used to take a team of scientists, lawyers, and analysts several weeks (finding the relevant documents, translating them, comparing them, and connecting the dots) can now be done by one person with AI tools in a matter of hours.  An investment committee should assume that competitors, partners, and rival bidders have already done it.

None of this means disclosure is bad.  Life science companies have to disclose.  They report to regulators, raise money from investors, work with partners, and file patent applications, and some disclosures, such as registering many clinical trials on ClinicalTrials.gov, are required by law.  What a buyer needs to know is whether the company managed its disclosures, whether it got something worthwhile in return for each one, and whether it still has an edge now that the market understands its technology.

The disclosure bargain has changed

Disclosure was never free.  Every patent application, paper, poster, trial registration, grant report, website update, conference talk, job posting, investor deck, and partnership announcement gives away a piece of the plan.  Competitors could always go looking.  But the work was slow and expensive, and the pieces tended to sit in separate silos, split by scientific discipline or by language.

Generative AI began to change that in late 2022.  The bigger shift came in late 2024 and early 2025, when several providers released “deep research” tools that can run a multi-step investigation on their own.  These tools can read hundreds of sources, pull the claims out of a patent family, compare molecules or methods, point out gaps, suggest design-arounds, and map who is collaborating with whom.  They produce a usable competitive brief in hours.  They also make mistakes.  But they are cheap enough that a competitor can run the analysis, check it, and run it again.  To be clear, AI does not make infringement, trade-secret theft, or breach of contract any more lawful than it was.  What it does is make lawful research into public sources, reverse engineering, and independent development much faster and cheaper.

Patent publication is where this bites hardest.  Most U.S. nonprovisional utility applications are published shortly after 18 months from the earliest filing date they claim.  A company can keep an application from publishing only in narrow cases.  The usual route is a nonpublication request, and that option disappears if the company files the same invention abroad or under the PCT, where it will publish at 18 months anyway.  Once an application publishes, it may teach much more than the claims that eventually issue will protect.  Biotech applications often stay in examination for years after they publish, so competitors get a long head start while the applicant is still finding out what protection it will get.  The law also pushes life science applicants to say more, not less: the written-description and enablement requirements of 35 U.S.C. § 112(a) reward more examples and more technical detail.  In practice, a company can be required to teach broadly and still end up with narrow claims.

Life sciences leaves an unusually rich public trail

Biotech and health-technology companies rarely disclose in one place.  They disclose in fragments.  A patent application may reveal targets, sequences, biomarkers, formulations, or model features.  A poster adds performance data.  A ClinicalTrials.gov entry lists endpoints, eligibility criteria, dosing, comparators, sites, and timelines.  A job posting hints at the modality, the manufacturing platform, or the regulatory plan.  A university page names the founding lab and the key scientists.  The marketing copy tells you which workflow problem the company thinks it has solved.

Put those fragments together and a competitor can make educated guesses about the target indication, how far along the program is, where the claims are likely to end up, how the company is validating its product, what it depends on, and what it will try next.  The competitor doesn’t have to copy anything to benefit.  It might time its own trial better, design a noninfringing alternative, file an improvement patent that blocks the target’s next step, recruit the target’s key people, or show up at a negotiation knowing exactly what the target depends on.

Why this matters to a buyer or sponsor.  A disclosure can create value for the whole field and still create very little for the company that made it.  When you buy a company, you are not paying for what it has already taught everyone else.  You are paying for rights you can enforce, know-how the company still controls, reliable access to data, the ability to execute, and a market position that others cannot copy quickly.

CRISPR and Amgen: two sides of the problem

CRISPR is a cautionary tale, though AI had nothing to do with it.  What it shows is how fast a foundational disclosure can crowd a field.  The key publications and patent filings came in 2012.  They revealed a platform with obvious follow-on potential, and several institutions raced to build on it.  More than a decade later, the fight over who first invented CRISPR-Cas9 genome editing in eukaryotic cells is still being litigated.  In May 2025, the Federal Circuit upheld the Patent Trial and Appeal Board’s ruling that the University of California group’s early applications lacked adequate written description.  It also threw out the Board’s conception analysis because the Board had applied the wrong legal standard, and it sent the case back.  On remand, in March 2026, the Board again awarded priority to the Broad Institute.  We don’t read this history as proof that publishing was a mistake.  We read it as a reminder that a breakthrough can split the value among many players, including whoever holds the foundational claims, the implementations, the improvements, the delivery systems, the manufacturing know-how, and the rights to particular applications.  Sorting out who owns what can take more than ten years.

Amgen v. Sanofi shows the drafting side of the same problem.  Amgen claimed a whole class of antibodies defined by what they do: bind to specified residues on PCSK9 and block PCSK9 from binding to LDL receptors.  The Supreme Court held, unanimously, that the claims were invalid for lack of enablement.  The specification did not teach skilled scientists how to make and use the full range of antibodies claimed without unreasonable experimentation.  The Court was careful to say that a patent does not have to describe every possible embodiment.  Still, the case highlights a real risk for investors.  A company can publish its examples, its screening methods, and its research roadmap, and still come away with claims much narrower than the platform story it has been telling the world and its investors.

The answer, when an invention can be reverse engineered and is worth protecting, is a better patent application, not a thinner one.  That means well-supported embodiments, real alternatives, fallback positions, and boundaries drawn on purpose.  It also means deciding which negative know-how (what the team tried that didn’t work) has to go into the application to support the claims, and which can stay a trade secret.  Defensive publication makes sense only when blocking others is worth giving up the secret.  And every filing should protect something the business actually values.  Portfolio size is not a goal in itself.

A clinical AI company can be mapped without special access

Take a fictional company we’ll call ReadmissionAI, which predicts which hospital patients are likely to be readmitted within 30 days.  Its website describes the problem and the performance it claims.  A health system’s press release says where the tool is being used.  Job postings reveal the data-engineering stack and the kinds of people it is hiring.  A conference poster describes the validation cohort and some of the model’s features.  A published patent application covers the model architecture, how data are normalized, and deployment options.  A study registration or quality-improvement report may describe endpoints and where the tool fits in the clinical workflow.

None of these items gives away the business.  Taken together, though, they may let a competitor with AI tools sketch a plausible map of the system.  It could compare the patent claims with the product on the market, spot features the claims don’t cover, work out where the data come from, find likely hospital partners, and try alternative models or workflows.  The map might be wrong.  But it costs so little to make that the competitor can keep refining it.  The patents may also offer less protection than they appear to.  Claims to predictive clinical software are vulnerable to subject-matter-eligibility challenges under 35 U.S.C. § 101, which can leave the real claim scope narrower, and less certain, than the product itself.

So ReadmissionAI’s defensible value has to come from things its public story doesn’t reveal.  The most important are exclusive or long-term rights to data, and curated labels linked to patient outcomes over time.  Next come documented data provenance and patient permissions, real integration into clinical workflow, and performance validated across multiple sites.  A clear regulatory position and quality system matter too.  So do server-side deployment, negative know-how the company has kept to itself, switching costs for customers, and contracts that limit reuse, model training, benchmarking, and competitive development.  These are much harder to rebuild from public sources, and they are what a buyer should be paying for.

The modern life science moat is an IP stack

We are not suggesting that companies stop publishing, partnering, or patenting.  We are suggesting that they treat each disclosure as an investment decision, and that they build several kinds of protection that overlap and reinforce each other.

• Start with secrecy.  Before disclosing any technical, clinical, operational, or commercial information, ask whether it should be a trade secret instead.  Identify the secrets that matter most.  Limit who can see them, keep a record of sensitive materials, train your people, and hold vendors to the same standard.  Keep evidence of those steps, because federal and state trade-secret law protects information only if the owner has taken “reasonable measures” to keep it secret.  Pay particular attention to negative know-how: failed experiments, process windows, assay conditions, data-cleaning rules, and manufacturing tolerances.
• Patent selectively and deliberately.  File where the right to exclude really matters and where you could detect and prove infringement, especially if the product can be reverse engineered.  Draft for the design-around a competitor is likely to try, not only for your preferred embodiment.  Line up publications, posters, fundraising, and partner discussions with your filing decisions, and ask whether a nonpublication request is available and makes sense.  Document what the human inventors contributed.  Only people can be inventors, and the USPTO’s November 2025 revised guidance treats AI as a tool, applying the traditional conception test to AI-assisted inventions.
• Update your contracts for AI.  A good NDA spells out what the recipient may use the information for.  Calling information “confidential” is not enough.  Deal with affiliates, residuals clauses (keep them narrow or leave them out), return and destruction, security, and remedies.  Bar the recipient from putting your confidential information into public or unapproved AI tools.  For approved tools, address data retention, model training, prompts, outputs, derivative models, audit rights, and ownership.
• Prove your rights in software and data.  Trace who owns the code, model weights, documentation, inventions, and improvements, through employees, contractors, universities, hospitals, vendors, and collaborators.  Make sure your data rights cover every use you plan: training, validation, regulatory submissions, commercialization, and transfer to a buyer.  Keep in mind that HIPAA de-identification addresses privacy.  It does not give you the contractual right to use the data or prove where the data came from.  Keep an up-to-date software bill of materials listing open-source components and third-party models.  FDA also requires one for many connected medical devices.
• Keep records of human authorship.  Copyright protects only work created by people.  The U.S. Copyright Office said so in its January 2025 report, and the courts said so in Thaler v. Perlmutter, which the Supreme Court declined to review in March 2026.  Log who wrote your code, documentation, databases, and content, and how.  Register your most valuable works, keep snapshots of source code, and use escrow or controlled deposits where needed.  You should be able to show what was created, by whom, on what terms, and with which dependencies.
• Make brand, technical controls, and commercialization work together.  Clear your key trademarks in the markets that matter.  Keep sensitive logic on your own servers when you can.  Tailor demos and disclosures to each audience.  Build regulatory expertise, integrations, supply and manufacturing relationships, a reimbursement strategy, and customer workflows that support your legal rights.  Each layer makes the others more valuable, and that is what makes a moat hard to cross.

What the investment committee should ask

1. If a competitor used AI to pull together every public disclosure about the target, what would it learn, and what could it build?
2. Which of the target’s advantages are still secret, hard to reproduce, protected by contract or technology, or backed by regulatory and commercial execution?
3. Do the patents protect the product and the obvious design-arounds, or do they mostly hand competitors a roadmap?
4. Can the target prove it owns, or has usable rights to, its data, software, models, improvements, clinical materials, and any contributions from universities or hospitals?
5. Do the target’s collaboration, evaluation, license, and customer agreements limit AI training, benchmarking, competitive use, and leaks through affiliates or vendors?
6. Does the target have a working trade-secret program, or just a confidentiality clause in its employment agreements?
7. Has the target’s own use of AI tools exposed confidential information or created doubt about who the inventors or authors are?
8. Will the IP transfer cleanly at closing, survive a change of control, and support the buyer’s commercialization plan?

Defensible enterprise value after disclosure

After decades of patent and life science practice, we would not tell a company to stop filing patents.  We would tell it to file for a business reason and to understand what each filing gives away.  And we would tell investors that patent counts, publication counts, and “proprietary AI” say little about how defensible a business is.

A better test is what the company still controls after it has disclosed.  We look for enforceable claims and protected know-how.  We look for clean rights to software and data, contracts that reflect how information actually moves today, and records showing who created and owns what.  And we look for technical and operational barriers, a trusted brand, and a commercial operation that turns all of this into lasting value for customers.  That combination is the modern moat.

For boards, strategic buyers, and sponsors, this work should start before a term sheet is signed.  First, map the public trail the way a competitor would.  Then ask whether the company’s nonpublic assets, legal rights, and operating capabilities hold up against what the public record has already revealed.  AI has made disclosure riskier.  It has also made a disciplined, layered IP strategy worth more than ever.

Selected authorities and practical resources

• Amgen Inc. v. Sanofi, 598 U.S. 594 (2023)
• Regents of the University of California v. Broad Institute, Inc., No. 22-1653 (Fed. Cir. May 12, 2025)
• Berkeley News, PTAB sides with Broad Institute on patent priority for CRISPR in eukaryotic cells (Mar. 26, 2026)
• USPTO MPEP § 1120, Eighteen-Month Publication of Patent Applications
• USPTO, Revised Inventorship Guidance for AI-Assisted Inventions, Fed. Reg. (Nov. 28, 2025)
• ClinicalTrials.gov Protocol Registration Data Element Definitions
• 18 U.S.C. § 1839, Trade Secret Definitions
• HHS Guidance on De-identification of Protected Health Information
• U.S. Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability (Jan. 2025)
• OpenAI, Introducing Deep Research (Feb. 2025)

This article is for general informational purposes only and does not constitute legal advice. The fictional ReadmissionAI example is an illustrative composite, not a description of an actual company. The views expressed are those of the authors and do not necessarily reflect the views of their firm or its clients.