Bankruptcy Issues Raised by Generative AI: Ownership, Licensing, and Asset Values

The rapid commercialization of generative artificial intelligence (AI) has created business models that depend heavily on data rights, intellectual property (IP) licensing, and cloud-based infrastructure, which implicate still-evolving regulatory protections and contractual restrictions. The same features that make AI companies scalable may make them uniquely complicated to restructure. Threshold questions over what, precisely, a debtor “owns” with respect to its AI systems and the kind of content being generated by them could have significant ramifications in a restructuring process.

What is Generative AI?

The term “generative AI” has all but morphed, in many respects, into the broader buzzword “AI.” Artificial intelligence, at an elemental level, is technology that attempts to mimic the human ability to analyze data, detect patterns, learn responses, and predict future behavior through algorithms and computational models.

Generative AI takes it a step further and focuses on creating new outputs (including text, images, and code) from existing datasets. Generative AI creations are testing the boundaries of IP protection and raising new questions about attorney-client privilege that courts are only just beginning to address. (For more information, read our prior articles Copyright and AI: Controlling Rights and Managing Risks and When AI Meets Privilege: Early Court Decisions.)

The Challenges of Restructuring AI Assets

Restructuring AI assets is likely to test the tenets of existing bankruptcy laws. The first thing routinely considered in a reorganization is what assets are available to satisfy a debtor’s obligations. The pool of assets is defined in section 541 of the Bankruptcy Code as the “estate.”

With AI assets, figuring out what is in the bankruptcy estate is not that simple. Section 541 of the Bankruptcy Code broadly defines the bankruptcy estate to include virtually all legal or equitable interests of the debtor as of the bankruptcy filing; however, a debtor’s AI assets may involve proprietary code and data integrated with components that are either licensed or dependent on third-party infrastructure, datasets, application programming interfaces (APIs), model weights, and service commitments. Content generated by such complex systems may therefore be subject to competing ownership claims or even infringement claims despite the debtor’s ownership of the data and algorithms that create the content. Even for content that is indisputably owned or licensed, whether such content is “intellectual property” under the Bankruptcy Code could impact two of a debtor’s key bankruptcy rights: assumption and assignment of contracts and sales of assets free and clear of liens, claims, and encumbrances.

Assumption and Assignment of Contracts

Section 365 of the Bankruptcy Code gives debtors a powerful restructuring tool: the ability (subject to exceptions) to assume and/or assign executory contracts that have value to the estate irrespective of contractual anti-assignment provisions and reject burdensome ones.

AI assets often involve complicated licensing structures. A typical technology licensing structure for an AI company may include

• upstream inbound licenses under which the debtor licenses technology, data, models, APIs, or software from another owner;
• outbound commercial licenses to partners or affiliates;
• customer-facing end-user licenses; and
• hybrid arrangements combining pure IP rights with hosted services, data access, and model governance obligations.

Early in 2026, a coalition of technology leaders launched the Shared AI License Foundation (SAIL) to facilitate shared access to AI patent rights for its members.[1] SAIL is designed to help member companies obtain non-exclusive licenses to foundational patent rights to speed development of new technologies and reduce legal costs and friction.

However, debtors relying on licensing arrangements that include non-exclusive IP licenses may encounter difficult restructuring choices under the Bankruptcy Code. Under section 365(c) of the Bankruptcy Code, contractual anti-assignment clauses, which generally are not enforceable in bankruptcy, continue to apply where applicable non-bankruptcy law excuses the non-debtor from accepting performance from a party other than the debtor. Courts have held that non-exclusive IP licenses may fall within this bankruptcy exception.

Moreover, if the debtor cannot assign the licenses, it also may not be able to assume them. There is currently a circuit split on whether an executory contract can be assumed by a debtor or debtor-in-possession without counterparty consent if it cannot be assigned:

• The Hypothetical Test: Bars assumption where applicable law would prohibit assignment without consent, regardless of whether assignment is contemplated. This approach is followed in the Third, Fourth, Ninth, and Eleventh Circuits.
• The Actual Test: Bars assumption only if assignment is actually intended. This approach is followed in the First and Fifth Circuits.
• The Footstar Test: Adopted by the Bankruptcy Court for the Southern District of New York, it allows a debtor-in-possession, but not a trustee, to assume (but not assign) an executory contract.

The approach of the governing jurisdiction should therefore be carefully considered before filing, because courts sometimes vary in how they apply the test to particular contract types.

Asset Sales

Lack of clarity over what a debtor “owns” can also have knock-on effects on other aspects of a debtor’s reorganization, including in the context of an asset sale. One of the most common restructuring tools in the Bankruptcy Code’s expansive toolbox is a “free and clear” asset sale under section 363, which allows companies to sell their assets “free and clear” of liens, claims, and encumbrances if certain conditions are met. This valuable right is used to monetize assets and grants debtors the ability to conduct sales expeditiously as compared to sales outside of bankruptcy.

The value of AI assets on the auction block is likely to turn on what the debtor owns and what a buyer can do with it. The distinction between assets that are “intellectual property” under nonbankruptcy law and those that aren’t, may have a big impact on asset value. Purely AI-generated content may not be protectable as intellectual property under copyright law absent sufficient human authorship, although works that incorporate AI-generated material may receive protection for human-authored elements.[2]

For patent law, the key issue is human inventorship and whether a natural person made the legally required inventive contribution. As the definition of “intellectual property” under the Bankruptcy Code is limited to the categories enumerated in section 101(35A)[3] and only to the extent protected by applicable non-bankruptcy law,[4] there is no reason to believe the bankruptcy courts will define IP assets more broadly than the federal courts.

In the absence of IP protection, the value of generative AI assets may be limited. Moreover, use of these assets by a purchaser may also be curtailed by regulations and data-privacy protections. These regulatory and contractual distinctions need to be carefully considered in connection with assessing the viability (and potential returns) of any sale process in this space.

Takeaways

The intersection of AI and bankruptcy law remains largely uncharted territory and so far there have been virtually no bankruptcy cases of AI assets to offer guidance on how courts will deal with these complex assets. For example, the recent Builder.ai case, In re Engineer.ai Corp., No. 25-10984-CTG (Bankr. D. Del. filed June 2, 2025), did little to advance the state of the law, as that case focused more on alleged fraud, business failure, and “AI washing” (i.e., exaggerated or fabricated AI capability) than on contested restructuring principles governing AI assets.

The nature of AI technology requires careful diligence and particular attention focused on the scope of the debtor’s estate, whether key assets qualify as protectable IP under applicable law, and whether key IP licenses are assumable and/or assignable under section 365(c) of the Bankruptcy Code. Ultimately, the next frontier of restructuring law may well be defined by a deceptively simple question: what, exactly, does an AI company own, and what can it transfer?