Want to learn more about drafting, negotiating, and understanding intellectual property and technology contracts and have 10 minutes to spare? Grab your morning coffee or afternoon tea and dig into our Tech Contract Quick Bytes—small servings of technical contract insights expertly prepared by our seasoned attorneys. This month, we explore the legal issues that arise when companies move AI from experimentation into commercial deployment.
Artificial intelligence (AI) contracting is changing. For many companies, the question is no longer whether they can experiment with AI. The harder question is how to move an AI application from a pilot into production, incorporate it into business operations, or sell it to customers.
That transition can expose issues that are easy to overlook when the technology is still experimental. Traditional SaaS, software development, data, and services agreements remain relevant, but AI raises additional questions about data use, intellectual property, model improvement, automated outputs, privacy, and responsibility when the technology does not perform as expected.
What follows are several issues companies should consider as they move toward AI commercialization.
Know What You Are Buying—or Selling
"AI" can describe very different technologies and commercial arrangements. A company may be licensing an AI-enabled SaaS product, accessing a model through an API, deploying an AI agent, obtaining implementation services, incorporating third-party AI into its own product, or commercializing a proprietary model it developed in-house.
The contract should accurately describe the technology and how it will be used, including relevant models, platforms, third-party components, implementation services, permitted users, deployment environments, and material dependencies.
For customer-facing or mission-critical uses, companies should also consider implementation milestones, acceptance criteria, service levels, human oversight, change management, and remedies if the AI functionality materially changes during the contract term.
AI Data Rights: Follow the Data
Data rights are frequently at the center of an AI transaction. The contract should distinguish among customer data, prompts, inputs, outputs, training data, usage information, and other information generated through use of the system.
One increasingly important question is whether the AI provider may use customer information to train or improve its models. The answer may vary, depending on whether the information is confidential, personal, proprietary, or commercially sensitive.
Companies should understand not only whether their data can be used, but what it can be used for, by whom, for how long, and for whose benefit.
Address IP Ownership and Licensing
AI complicates traditional intellectual property provisions because multiple layers of technology and content may be involved.
Contracts should address ownership of preexisting technology, custom developments, prompts, outputs, configurations, fine-tuned models, and improvements. If a customer expects to commercialize AI-generated outputs or incorporate them into its own products, the agreement should provide sufficient rights to do so.
The parties should also consider third-party IP risk. What representations does the provider make regarding its models and training materials? What happens if an output allegedly infringes another party's rights? Traditional software IP indemnities may need adjustment to address AI-specific risks and exclusions.
Don't Treat Privacy and Security as Separate Issues
AI systems can process significant amounts of personal, confidential, or sensitive information. Privacy and security therefore need to be integrated into the commercial arrangement.
Companies should determine what data the AI system receives, where it is processed, whether subprocessors or additional model providers are involved, how long information is retained, and whether it can be reused. Existing DPAs and security exhibits should be reviewed to ensure they adequately address the AI use case.
For AI agents or automated systems capable of taking actions rather than merely generating content, access controls and system permissions deserve particular attention.
Allocate Risk in AI Contracts
AI performance is probabilistic, and outputs can be inaccurate or unpredictable. Agreements should address warranties, disclaimers, human review requirements, prohibited uses, compliance responsibilities, indemnification, liability caps, and available remedies.
The appropriate allocation will depend on the use case. An internal productivity tool presents a risk profile that is different from that of an AI system interacting directly with customers or making consequential recommendations.
Companies should also plan for termination. Can data and configurations be retrieved? Can an AI-enabled service be transitioned to another provider? What happens to trained or customized functionality? Commercialization means thinking about the entire life cycle—not simply getting the pilot launched.
AI contracting remains technology contracting at its core, but the underlying rights and risks are becoming more complicated. Companies moving AI into production should treat data rights, IP ownership, privacy, operational controls, and liability allocation as part of the commercial architecture—not as issues to resolve after the technology has been selected.
*****
If you or your company would like to discuss any of these AI commercialization or technology contracting issues, please contact A.J. Zottola.
To receive more Tech Contract Quick Bytes, be sure to subscribe. Click here to learn more about Venable's IP Tech Transactions services. Looking for tech contract support? Our Contract Concierge provides clients with access to a dedicated team of Venable's experienced tech, IP, and privacy attorneys to assist with contract demands, drafting, and negotiation.