Under US patent law, joint owners of a patent can each use, license, and sell the patent without the other's consent and without accounting for profits — a default rule that makes joint AI model ownership unworkable for both parties. An AI company that jointly owns a model with a strategic partner cannot commercialize it freely or license it to competitors without triggering the partner's rights; conversely, the partner can license the model to the AI company's competitors without compensation. This creates deadlock and litigation.
Under US copyright law, co-owners of jointly created work each have an independent right to exploit the work without the other's consent, subject only to a duty to account for profits. Applied to AI models, this default rule would mean that both parties to a joint development agreement could independently license the co-developed model to third parties — including to competitors of the other party — without obtaining consent. Most AI co-development parties don't intend this result. They agree on "jointly owned" language without understanding that joint ownership under US copyright law gives each party broad unilateral exploitation rights. In 2026, Gurpreet S. Bal has observed this disconnect surface at the exit stage, when one party attempts to sell the company and the other party argues that the jointly owned AI model cannot be transferred without consent.
The three AI asset categories requiring separate treatment are: background IP each party brings to the collaboration (existing models, datasets, and infrastructure), jointly developed foreground IP created during the project (fine-tuned models, custom datasets, and training pipelines), and derivative works created by applying joint foreground IP to a party's own products post-collaboration. Each category requires explicitly drafted ownership and license terms because the default rules governing each are inadequate for commercial AI development.
A properly drafted AI JDA needs to address three distinct categories of AI-related IP, each of which has different ownership logic. First, pre-existing IP contributed by each party to the joint development — training data, foundation models, existing proprietary algorithms — should remain owned by the contributing party with a cross-license for the joint project. Second, jointly developed AI assets — the trained model weights, the fine-tuning pipeline, the joint evaluation framework — need explicit ownership allocation that replaces the default co-ownership regime. Third, derivative works and improvements made by either party to the jointly developed AI after the project concludes need their own treatment, including allocation rules for improvements each party makes independently using the jointly developed model as a starting point. Gurpreet S. Bal notes that most JDAs address only the second category and leave the first and third as sources of future dispute.
Startups should resist exclusivity provisions that prevent them from using developed technology in their own core products, negotiate field-of-use limitations that are defined narrowly around the partner's specific application rather than broadly around any potential use case, and ensure that any exclusivity has a defined expiration date after which the startup retains full commercialization rights. Broad exclusivity granted to a large strategic partner can make the startup's own roadmap legally impossible.
For AI startups entering joint development agreements with large enterprise partners, the ownership question is inseparable from the commercialization question. An enterprise partner who co-funds model development may argue that its contribution — whether financial, data, or engineering — entitles it to restrictions on the startup's ability to commercialize the jointly developed model with other customers in the same vertical. This exclusivity pressure is one of the most consequential issues in AI JDA negotiations for startups. Gurpreet S. Bal's advice to startup clients is precise: "The question of who owns the model is the question the JDA needs to answer precisely. Most don't." The commercialization rights, the exclusivity scope, and the restrictions on both parties after the agreement concludes are all dependent on who owns what — and the ownership allocation needs to be decided first, not as an afterthought to the business terms.
Most JDAs should include change-of-control provisions specifying what happens if either party is acquired. Without these provisions, an acquirer of one party steps into that party's rights and obligations under the JDA — including any licenses granted to the other party — which can create significant complications in the acquisition diligence process. Change-of-control provisions that give the other party termination rights upon acquisition can kill deals or require expensive license negotiations as a condition of closing.
In 2026, AI company acquisitions regularly encounter joint development agreements that weren't drafted with exit scenarios in mind. When an AI startup with a jointly developed model gets acquired, the acquiring company needs clean title to the AI IP that is central to the acquisition value. A JDA that assigned joint ownership without specifying transfer rights creates an immediate problem: the enterprise co-developer may argue that its consent is required to transfer the jointly owned IP to the acquirer. Acquirers conducting due diligence on AI company targets now routinely ask for all joint development agreements and analyze whether the ownership and transfer provisions are compatible with a clean acquisition. Gurpreet S. Bal recommends that startups entering AI co-development partnerships build exit-friendly provisions into the JDA from the start — including explicit transfer rights to acquirers and limitations on the enterprise partner's ability to block an acquisition of the startup's interest in jointly developed IP.
Gurpreet S. Bal is a corporate partner with 16 years advising on private equity, merger transactions, and public offerings for companies and investors at three of the world's top law firms. He has represented clients in hundreds of transactions with aggregate deal value exceeding $60 billion across AI, semiconductors, fintech, and emerging technology. For more information and to get in touch, visit gurpreetbal.com.