Background

Artificial Intelligence (“AI”) has been utilized by innovative companies across the array of industry spaces, but its impact on biotech (and human health in general) could become the most significant among all sectors.

AI is being leveraged across the drug development spectrum to accelerate inherently burdensome tasks including the identification of targets and leads, optimization of leads, and prediction of preclinical and clinical safety and efficacy.[1]

With respect to target identification in particular, AI models trained on multiomics, literature, clinical phenotypes, and other data are identifying candidate disease biology hypotheses and entirely new protein and pathway targets.[2] For instance, some AI systems have been reported to infer receptor-ligand interactions based on correlated gene expression signatures, even absent prior structural annotation.[3]

With regard to identifying and optimizing lead compounds, some generative AI (for example, graph neural networks, diffusion models, and molecular design systems) that are fed with an array of data sets have been reported to propose novel chemical entities predicted to, for example, bind to a selected target.[4] With composition of matter patents being highly valuable, with some capable of protecting entire therapeutic platforms alone, new chemical compound outputs in particular can represent the most commercially valuable pieces of the pipeline.

It is therefore no surprise that numerous companies have emerged that have created AI drug discovery platforms to (1) serve companies in drug development, and, in some instances, (2) build and advance their own pipelines.[5] 

While groundbreaking and exciting, these approaches can also trigger acute patentability issues relating to inventorship, obviousness, and enablement, as well as practical data management and confidentiality issues. As a result, market players and their counsel face a new dynamic in collaboration, licensing and other deals relating to how to properly account for the risks and benefits presented by the use of AI as a prominent drug discovery tool.

Hypothetical Deal Scenario

For purposes of this post, we consider the following deal example.

Company A is an AI-platform company that has an existing compound library that was created through the utilization of its platform. As part of its business model, Company A is seeking to leverage its compound library for licensing deals. Company A is also offering its AI platform as a service to third parties for the discovery of new drugs.

Pharma Company B is a pharmaceutical company that is seeking to license certain of Company A’s compounds in its library as well as Company A’s AI platform to discover new compounds for a long unaddressed oncology indication (“Indication X”).

Interests Analysis and General Deal Framework

In this scenario, Company A’s interests are in (1) maximizing the value of its licensed compounds in its library and (2) leveraging the AI platform in a manner that results in significant service revenues while protecting its core background IP and potentially new IP.

On the other side, Company B’s interests are in obtaining exclusive and/or ownership rights in the existing and to-be-discovered compounds to support development and commercialization of drugs directed toward addressing Indication X.

Here, the deal framework would be structured as a classic compound license, assuming that Company A has properly protected its core compounds with patents. Company B will certainly evaluate the coverage and value of the patent portfolio before entering into a deal.

The framework would likely also include a service component for utilization of the platform, raising negotiation points relating to ownership over resulting IP and data. The provisions would also address exclusivity, IP ownership, diligence, data, confidentiality, and other key items.

Library Compounds

General deal terms for the compound license would include upfront licensing fees, milestone payments correlating to (for example) regulatory progress, and royalties and sales milestone payments tied to net sales of the resulting commercial products covered by the applicable patents.

Because it would want full rights to commercially exploit the patented compounds, Company B will likely seek an exclusive license, and not take a non-exclusive license, for the existing compounds in Company A’s library.

If offering exclusive rights, Company A may be well-served to segment the field of use into a niche field (for example, a very specific target or disease indication) that it is comfortable not granting rights to others in. This leaves the opportunity to license the compounds in other fields even on an exclusive basis. Offering exclusivity will also serve as grounds for increasing the price of the license.

Platform-Generated Compounds

With respect to utilizing the platform as a service, Company A will have strong interest in maximizing revenues for the service and potentially owning or reserving significant rights in the IP that is generated therefrom.

In our scenario, Company B may seek to own the new compounds outright, particularly with the development and commercialization program being led by Company B. In that case, Company B’s objective would be to secure ownership of all compounds identified or optimized using the platform, with Company A compensated through service fees. Company A may also seek additional fees based on development and commercialization milestones and/or downstream royalties.

On the other hand, if, for example, the class of compounds in Company A’s library has applicability in other indication areas beyond Indication X that Company A feels have commercial value, Company A may seek to retain ownership over the IP rights in the new compounds in the subject class, but agree to fold those additional compounds into the exclusive license for compounds solely for Indication X.

In that situation, with ownership of the IP covering the new compounds vesting with Company A, Company B may seek to drive down or eliminate the cost for the service. This would be based upon the fact that the resulting compounds will be owned by Company A and added to the licensed library compounds and should therefore be treated the same or similar to the existing library compounds. This arrangement would allow Company B to pursue its program for Indication X under an exclusive license while allowing Company A to preserve its right to license or otherwise utilize the new compounds in other indication areas as it is doing with its existing library compounds.

Furthermore, with everything being negotiable, Company B could also seek to purchase ownership rights to the entire class of compounds old and new. This would certainly simplify – and significantly raise the value – of the deal.

The parties could also agree to joint ownership rights in the compounds, but that arrangement could complicate matters including with regard to attracting investment and risking that commercialization exclusivity does not vest with a single party.

Inventorship Considerations

The laws of inventorship relating to AI are in flux.  In at least the U.S., a human being (and not AI) must be an inventor of patent claims and (based on recent USPTO policy guidance) at least one human being must provide a “significant contribution” to the conception of the claimed invention.[6]

What is meant by “significant contribution” is subjective and based on a judge-made analysis applied by courts to joint inventorship scenarios for a variety of inventions and technological areas over the decades. While the Federal Circuit has held that “only a natural person can be an inventor, so AI cannot be,”[7] courts have not yet applied the “significant contribution” standard to AI-assisted inventions.

How courts will deal with inventorship of AI-developed inventions is therefore uncertain. The consequence of patents that are issued having inventors that were not properly named and with no other human being having made a “significant contribution” is that the patent could be found invalid in a district court action.

Based on this, in our scenario, Company B may seek representations and warranties from Company A that the inventors of the licensed patents are properly named and that at least one inventor  provided a “significant contribution” to the conception of the invention of each claim in the licensed patents. Company B would also include standard language that would cause the agreement and royalty obligations to terminate upon a final judgment of invalidity of applicable patents, which would at least mitigate the obligation of ongoing royalties.

Ownership and Use of Data

In our scenario, Company A would own the AI platform and likely any improvements thereof. However, the agreement must clearly define who owns the data that is utilized or produced by the AI platform, including (1) data that is necessary and useful in the subject regulatory process in obtaining approval for the drug used to address Indication X; and (2) data that may be needed post-approval.

AI systems can generate information relating to, for example, predicted binding scores and toxicity risk models, as well as in silico simulations, and other key data that can be relevant to the regulatory process. Recent FDA guidance provides policies on the use of such data to support regulatory decisions on drug safety, effectiveness, and quality, indicating that the party who pursues regulatory approval would need such data to successfully proceed.[8]

AI systems may also be needed post-approval and during commercialization, for example to perform companion diagnostics by identifying patients who are likely to benefit from treatment,  recommending treatment regimens for personalized care, predicting the effectiveness of treatment, and monitor patient responses.

Therefore, in our deal, Company B, the party that would pursue regulatory approval and commercialization, will want to include language that requires that it (1) either solely owns, jointly owns, or otherwise has perpetual, irrevocable, and worldwide rights to access and utilize data that is necessary and useful for regulatory purposes; (2) is provided a physical copy of all such data which is updated on a regular basis during at least the agreement term; and (3) has perpetual, irrevocable, and worldwide rights to utilize the AI platform post-approval in situations where the platform is needed for commercialization (for example, companion diagnostics). This way, Company B would ideally have both possession of and rights to the data and have access to the AI platform for necessary services following the natural expiration or earlier termination of the agreement.

Ownership and Confidentiality of Training Data

Of course, the parties must determine what training data will be used to develop the new compounds, who owns that training data, who has access to that training data, and how it is handled.

If, for example, Company B owns the training data that will be utilized in Company A’s platform to develop the new compounds, Company B should seek to make sure that the confidentiality of the data is maintained and access to it is limited.

This can be very tricky given that Company A’s platform will doubtless learn from and retain copies of the data due to the nature of the model itself.  Company B may seek to require that Company A not have access to the raw training data and only be given strict limited access to, for example, aggregate summary metrics of the training data, if that is even possible from a technical perspective.

Company B may also utilize federated learning (decentralized AI training methods that store data on disparate storage mediums) or differential privacy (mathematical techniques that inject statistical noise into data before feeding it into an AI system) to mitigate the risk of disclosure.[9]

Company B may also seek to propose provisions that require that Company A will maintain the training data as confidential for an unlimited term and not access, use, distribute, copy, reproduce, or reverse engineer the training data or related know-how. This includes restrictions preventing Company A from using the training data in its platform for other clients.

Significant hurdles will likely exist in such regulation of such use of the data because training data may be inextricable intertwined with the platform that it teaches. Regardless, in this scenario, Company B should be vigilant in putting provisions in place, or requiring Company A to take burdensome technical steps, if Company B wants to properly maintain confidentiality of its training data.

Because regulators such as the FDA may scrutinize AI-derived data, the agreement should also require that the AI platform’s operation and validation conform to applicable regulatory guidance. The agreement can specify documentation standards that support model transparency, validation, and reproducibility, aligning with emerging FDA draft guidance on AI credibility and lifecycle oversight.

Finally, the parties should also, where applicable, define the scope of data rights and privacy obligations in detail, including how patient-related datasets are collected, stored, and shared. These provisions should reference applicable frameworks, such as the Health Insurance Portability and Accountability Act (HIPAA) of 1996.

Indemnification and Risk Allocation

In addition to confidentiality and data ownership provisions, parties should carefully negotiate indemnification obligations relating to the use of AI systems and associated data. Given the novelty of AI-driven discovery platforms, the risks extend beyond conventional IP infringement to encompass the integrity, provenance and third party rights relating to the data used to train such models.

For example, Company B will likely require indemnification from Company A for any claims arising from the use of training data for the base platform in which Company A lacked appropriate third-party permissions or licenses. This includes, for instance, situations in which confidential, copyrighted, protected patient, or otherwise restricted third-party data was used to train the AI system without authorization, leading to potential claims of data misuse or misappropriation. Company A, in turn, may require reciprocal indemnification from Company B where Company B contributes training data that has similar attendant issues.

Similarly, Company B may seek indemnification for damages or losses caused by malicious code, technical failure, or unauthorized access (for example, hacking events) associated with Company A’s platform. Company A, conversely, may seek to limit its exposure by limiting liability to service fees or excluding special or indirect damages. If Company B is allowed access to itself utilize the platform, Company A may seek indemnification for claims relating to misuse of the platform.

To strengthen these indemnification provisions, Company B may also require that Company A implement and maintain specific technical and organizational measures, such as encryption standards, penetration testing, sequestered storage conditions (e.g., on-prem storage), and disaster recovery plans, with contractual certification or audit rights to confirm compliance. This approach reflects a growing recognition that AI-based collaborations require both robust technical safeguards and clear contractual allocation of risk for data provenance and platform reliability.

Of course, each party may seek indemnification for claims of third party infringement and require reps and warranties that there is none.  Given the potential for significant exposure in the event of a third party infringement claim, whether and to what extent each party agrees to these clauses will likely be a key point of negotiation.  

Patent Prosecution for New Compounds

The control of, and payment for, patent preparation and prosecution for any new compounds discovered using the AI platform will likely turn on which party owns the underlying intellectual property. If Company B owns patent rights to the new compounds, Company B would have the first right to control preparation and prosecution of any related patent applications, at its own expense.

In that scenario, Company A may request a right to review filings to at least ensure that its background IP, including any proprietary aspects of the AI platform, is not inadvertently disclosed or encumbered, or that the patent filings do not conflict with any of Company A’s other programs. Company A may also seek rights to own, prepare, and prosecute patent applications in the event that Company B opts not to prosecute.

Conversely, if Company A retains ownership of the compounds, it may seek to have control over patent prosecution, although, as a party with key commercialization rights, Company B would be expected to have significant collaborative rights, particularly with respect to claim scope and geographic coverage, to align with Company B’s commercial objectives for Indication X.

This cooperative model could mirror traditional research collaboration agreements, with Company A leading prosecution and the parties coordinating strategy on application drafting and office action responses,  and, in our scenario, confirming inventorship determinations where humans and AI provide input to the claimed invention.

Bankruptcy Considerations

Given that some AI-platform companies are startups and may not have proper funding, Company B should also take careful consideration of the solvency of Company A into account, and how insolvency and bankruptcy (for example) may affect the deal.

If Company B is building its program for Indication X on the utilization of Company A’s platform, including the compounds and data that it generates, as discussed above, Company B should ensure that its regulatory and commercial program can continue post-agreement where applicable.

In circumstances where the AI platform is required post-regulatory, for example, as discussed above for companion diagnostics, Company B will want to seek to make sure that it can continue to utilize the platform after conclusion of the agreement including in the event of bankruptcy of Company A.

For example, Company B could seek provisions requiring Company A to place a functioning and regularly updated copy of the platform into escrow during Company B’s commercialization including after the core agreement concludes. In that situation, an escrow agent would be charged to ensure that Company B has rights to at least access or utilize the platform for use with the required companion diagnostics.

The exact terms of this provision should be carefully crafted and could be difficult to negotiate given that Company A would likely consider much of its platform – including its underlying code – to contain trade secrets. Regardless, Company A would need to address this concern because it is possible that Company B would not be able to sell its product without access to Company A’s platform and Company B will need absolute assurance that it has such access.

Confidentiality and Post-Termination Handling of AI Systems

In addition to bankruptcy-related protections, agreements involving AI platforms should address how confidential data, trained models, and platform derivatives are handled upon expiration or termination of the collaboration. Where Company A builds or customizes an AI model specifically for Company B’s application (for example, directed discovery of compounds for Indication X), Company B may insist that the model and all associated outputs be segregated or “sequestered” from Company A’s general platform architecture. This ensures that the learnings or parameters derived from Company B’s proprietary data are not inadvertently retained in or reused by Company A’s system for other clients.

The parties may therefore include provisions requiring Company A to maintain such project-specific models on dedicated infrastructure during the term and to delete, anonymize, or otherwise destroy the relevant models and datasets following termination, except to the extent retention is required by law or expressly permitted for audit or regulatory purposes. In practice, complete deletion can be technically complex, as model weights may encode statistical representations of the training data. As a result, contracts often include a hybrid obligation: to render retained model artifacts unusable for future training or inference relating to other programs. This process would also likely be time consuming and expensive, but the sensitivity of the information may justify taking these steps.

Company B may further seek post-termination rights to obtain a static, escrowed copy of the trained model and relevant outputs for archival, regulatory, or defensive purposes. Where Company A intends to reuse general platform components (for example, code or algorithms not specific to Indication X), the agreement should clearly define which elements constitute background technology versus project-specific deliverables. These distinctions will help prevent disputes over whether continued use of an AI model by Company A constitutes misuse of Company B’s confidential information.

Conclusion

Licensing counsel must carefully consider the facts and circumstances in deals involving AI used as a drug discovery tool. Specific attention should be given to the scope of the license grant, service parameters, related fees, inventorship risks, ownership of regulatory and training data, confidentiality of training data, and other items discussed above. Other considerations should also be taken into account given the unique nature of AI platforms as applied to each specific deal situation.

[1] https://www.mckinsey.com/industries/life-sciences/our-insights/ai-in-biopharma-research-a-time-to-focus-and-scale

[2] Ocana, Alberto, et al. “Integrating artificial intelligence in drug discovery and early drug development: a transformative approach.” Biomarker Research 13.1 (2025): 45; Wang, Joanna. “Navigating the USPTO’s AI inventorship guidance in AI-driven drug discovery.” Journal of law and the biosciences vol. 12,2 lsaf014. 2 Aug. 2025, doi:10.1093/jlb/lsaf014; see also https://www.drugtargetreview.com/news/166597/ai-platform-detects-new-drug-targets-in-minutes/

[3] Vahid, Milad R., et al. “DiSiR: fast and robust method to identify ligand–receptor interactions at subunit level from single-cell RNA-sequencing data.” NAR Genomics and Bioinformatics 5.1 (2023): lqad030.

[4] Adegbola, Itunuoluwa, Integrating Generative AI with High-Throughput Screening for Accelerated Drug Discovery (June 20, 2025). Available at SSRN: https://ssrn.com/abstract=5317101; Boston Consulting Group & Wellcome Trust, Unlocking the potential of AI in Drug Discovery, (2023) https://wellcome.org/sites/default/files/2023-06/unlocking-the-potential-of-AI-in-drug-discovery_report.pdf 

[5] For example, systems like Insilico’s Chemistry42 generate molecules which are then synthesized and tested in vitro. Ivanenkov, Yan A., et al. “Chemistry42: an AI-driven platform for molecular design and optimization.” Journal of chemical information and modeling 63.3 (2023): 695-701.

[6] https://www.jdsupra.com/legalnews/key-inventorship-considerations-in-ai-6047492/; “Inventorship Guidance for AI-assisted Inventions,” PTO-P-2023-0043, available at https://www.federalregister.gov/documents/2024/02/13/2024-02623/inventorship-guidance-for-ai-assisted-inventions (the Guidance) (emphasis added; https://www.rothwellfigg.com/publication-the-usptos-inventorship-guidance-for-ai-assisted-inventions;

[7] Thaler v. Vidal, 43 F.4th 1207, 1213 (Fed. Cir. 2022), cert denied, 143 S. Ct. 1783 (2023).

[8] https://www.biosimilarsip.com/2025/07/23/beyond-guinea-pigs-patent-risks-and-opportunities-in-ai-enabled-drug-development/; https://www.biosimilarsip.com/2025/02/13/from-algorithms-to-approvals-navigating-ai-in-drug-and-biological-product-regulation/

[9] Kint, Stefan, Wilfred Dolfsma, and Daniela Robinson. “Strategic partnerships for AI-driven drug discovery: The role of relational dynamics.” Drug Discovery Today 29.12 (2024): 104242.

Disclaimer: The information contained in this posting does not, and is not intended to, constitute legal advice or express any opinion to be relied upon legally, for investment purposes or otherwise. If you would like to obtain legal advice relating to the subject matter addressed in this posting, please consult with your attorney. The information in this post is also based upon publicly available information, presents opinions, and does not represent in any way whatsoever the opinions or official positions of the entities or individuals referenced herein.