AI and IP in IBM Contracts
Artificial intelligence contracts present new risks around data use, intellectual property (IP) ownership, and liability. When using IBM’s AI platforms, such as Watson and WatsonX, enterprises must be vigilant.
Without the right protections, sensitive data could be repurposed, ownership of custom AI models might be unclear, and responsibility for AI-generated mistakes could fall on the customer. Read our ultimate guide to IBM AI & Analytics Licensing: Watson, Watsonx, Cognos, SPSS, and More.
This guide outlines how to secure your data, models, and rights when contracting with IBM for AI solutions. We focus on key areas—data usage, output ownership, and indemnity—to help legal, IT, and procurement leaders negotiate contracts that protect their interests.
1. Data Usage by IBM (Watson data usage rights)
IBM’s standard cloud terms often permit the company to use certain customer data to improve services in an aggregate manner. For example, in some Watson services, non-Enterprise plans by default allow IBM to log and learn from interactions to refine their AI models.
This means if you use Watson Assistant or similar services on a basic plan, your transcripts or prompts might be analyzed by IBM to enhance Watson’s general capabilities.
The risk: Without restrictions, sensitive business data could be used outside your specific instance, even if anonymized, contributing to IBM’s broader AI training or analytics.
Fortunately, IBM provides ways to limit data usage. Opt-out mechanisms exist: Watson services allow customers to disable IBM’s learning from their data (for instance, setting an X-Watson-Learning-Opt-Out:true header on API calls, or choosing Premium plans where data is isolated by default).
In Enterprise and Dedicated plans, IBM typically refrains from using your content for service improvements unless you opt in. However, these protections should not be assumed; they should be negotiated explicitly in the contract.
Insist on a clause stating that IBM will only use Customer Data to deliver the service to you and not to train or improve any general AI models or services without your consent.
This clause ensures that any data or prompts you input into Watson or watsonx remain within your control and are not mined to benefit IBM’s overall product offering.
Additionally, clarify that IBM may not derive or share “unique insights” from your data with others. IBM’s public principles claim “your data is yours alone”, so hold them to it in writing.
An ideal contract should state that IBM cannot use or disclose your content or any insights derived from it beyond providing the service to you.
By tightening data usage rights, you eliminate the gray area that might allow IBM to use your information for AI training, aggregate analytics, or product development.
In summary, make data usage restrictions a non-negotiable term: your enterprise data should only fuel your solutions, not IBM’s next AI model.
2. Output & Model Ownership (AI output ownership)
Who owns the outputs and models created using IBM’s AI services?
By default, IBM retains ownership of its base software and any pre-trained models it provides. However, the new AI assets generated from your data—chatbot responses, analysis results, fine-tuned models—should belong to you.
IBM’s contracts often specify that the client retains ownership of their “Content,” which includes input data and resulting outputs.
In practice, this means you don’t automatically surrender IP rights to the text, images, or predictions Watson produces for your organization. Nonetheless, it’s crucial to spell this out: the customer owns all outputs and any custom-trained models derived from their data.
IBM’s Watsonx platform allows you to train or fine-tune AI models on your proprietary data. Ensure your agreement says that any model weights or configurations generated using your data are your intellectual property (or at least exclusively licensed to you).
Without clear terms, there’s a risk IBM could argue it has some rights to the improved model or could reuse learnings from your model elsewhere.
While IBM publicly promises that “you own the algorithms” you build with Watson, only a written clause guarantees it. Negotiate language such as: “Customer retains exclusive ownership of any models, model customizations, and outputs developed from Customer Data, and IBM shall not use or access such models or outputs for any purpose other than providing the service to Customer.”
Also, be aware of any generic improvements clause. IBM may include wording that it owns or can use “residuals” or improvements to its platform that arise from providing the service. It’s fine if IBM improves its base software over time, but make sure they can’t claim ownership of your specific solution.
A balanced approach is to grant IBM rights to its underlying technology and any general enhancements, but not to your specific model configurations or data-trained insights.
For example, if your team trains a financial forecasting model on Watsonx using IBM’s base model plus your data, the contract should confirm that the tuned model is for your exclusive use.
Some enterprises even negotiate that IBM deliver the trained model files at contract end, or delete them, to prevent any reuse. In summary, secure your AI output IP: IBM’s tools, but your results, and your rules on how they’re used.
3. Indemnity for AI Outputs (AI IP clause IBM)
AI-generated content can inadvertently infringe on the intellectual property of third parties.
For instance, a Watsonx-generated marketing blurb might unknowingly plagiarize copyrighted text, or a code snippet from an IBM AI coding assistant might resemble someone else’s proprietary code.
Who bears responsibility if your company is accused of IP infringement due to AI output? Without special terms, the default answer is usually you, the customer.
Historically, many providers (IBM included) disclaimed liability for what the AI generates. The liability shift meant if a third party sued over an AI output, the customer would have to handle it alone.
However, the industry is changing. Competitors like Microsoft, Google, and AWS have introduced enterprise indemnification for AI outputs—essentially promising to defend and financially cover customers if an IP infringement claim arises from using the vendor’s AI tools. IBM has followed suit, but only in specific contexts.
Notably, IBM now extends its standard IP indemnification to the outputs of its own foundation models on watsonx.
This means if you use IBM’s proprietary models (such as the Granite series in watsonx.ai) and, say, a generated output triggers a copyright claim, IBM would step in to defend and indemnify your company. IBM has even touted that it does not cap its indemnification liability for these IBM-developed models, which shows a strong commitment.
The catch:
This protection may not cover all scenarios automatically. If you use third-party or open-source models via IBM’s platform, or if you intentionally misuse the AI (e.g., prompting it to recreate a famous copyrighted text), indemnity might not apply. And unless it’s explicitly in your contract, you’ll want to ensure this promise is on paper.
Negotiate an indemnity clause specific to AI outputs. It should state that IBM will defend and hold the customer harmless against any third-party IP claims arising from the content generated by the AI services, ideally including an obligation to cover any settlements or damages.
Also, crucially, ensure that IBM waives any requirement that you indemnify them for simply using their AI. (IBM’s policy with its models is to not ask customers to indemnify IBM for IP issues, which is good—just make sure that’s reflected in your terms as well.)
When negotiating, use your leverage. Point out that other cloud providers offer similar or better indemnities for AI. If Microsoft will cover your costs in an AI lawsuit, you expect IBM to at least match that.
Enterprise clients have successfully negotiated stronger indemnity clauses with IBM that go beyond the boilerplate. If IBM is unwilling, consider it a red flag and weigh the risk carefully.
In the end, the goal is not to be left holding the bag for an AI’s mistakes. An ideal IBM contract will explicitly shift that risk off your shoulders, so you can innovate with Watsonx confidently, knowing IBM will back you if something goes wrong.
(Table: AI Output Indemnity – IBM vs. Key Competitors)
| Provider | Indemnity for AI Outputs? | Notes |
|---|---|---|
| IBM | Yes (Negotiable) – Offers IP indemnity for outputs of IBM-developed models on watsonx; must be in contract. | IBM promises uncapped indemnity on its own foundation models. Ensure coverage extends to your use case. |
| Microsoft | Yes – “Copilot Copyright Commitment” covers enterprise users of its AI (with conditions). | Will defend and pay for copyright claims from use of Microsoft’s AI copilots, assuming responsible use. |
| Yes – Indemnifies outputs from select generative AI services for business customers. | Similar coverage for IP claims, not applicable if output is used maliciously or to intentionally infringe. | |
| AWS | Yes – Offers uncapped indemnity for generative output of its proprietary models (e.g., Amazon Titan). | Only applies to Amazon’s own models on services like Bedrock; third-party model outputs may be excluded. |
Use competitor commitments as a benchmark when negotiating with IBM.
4. Checklist of AI Contract Must-Haves
When reviewing or negotiating an AI contract with IBM, ensure it includes the following key clauses and protections:
- No Use of Customer Data for Broad Training: The contract should state that IBM may only use your data to provide services to you. Any use of your prompts, datasets, or logs to train or improve general Watson/AI models is prohibited (or requires your explicit, written permission).
- Exclusive Ownership of Outputs and Models: It must be clear that you own all AI outputs, custom models, and configurations created from your data. IBM retains its base tools, but anything produced for your organization (from chatbot answers to fine-tuned model weights) is your intellectual property.
- IBM Indemnification for AI Outputs: Include a strong indemnity clause where IBM agrees to defend and cover you if third parties claim an AI-generated output infringes their IP. This should cover legal defense and any settlements or damages from such claims.
- Data Residency and Sovereignty: For regulated data, the contract should commit that your data (and any AI outputs) will be stored and processed only in approved regions or environments. If you require an EU-only (or other jurisdiction) solution, ensure IBM contractually guarantees in-region data handling and compliance with local data laws.
- Exit and Portability Rights: Negotiate an exit clause that ensures you can retrieve all your data and trained models upon contract termination. IBM should assist in providing model files, data exports, and documentation so you can port your AI solution elsewhere or continue it on your own infrastructure. This clause protects you from vendor lock-in and data loss.
(Use this checklist during contract reviews to spot gaps and add necessary protections.)
5. FAQs
Q: Will IBM use my company’s data to train Watson models?
A: Not if you negotiate restrictions into your contract. By default, some IBM services might use customer data for product improvement, but you can opt out. Always include a clause banning IBM from using your data to train general models.
Q: Who owns a model trained in Watsonx?
A: It should be you. Out of the box, IBM provides the tools but may retain rights to its base technology. You need to explicitly state in the contract that any custom-trained Watsonx model (using your data) is your property and under your control.
Q: Can IBM reuse my AI outputs to train others’ models?
A: Not if your contract specifies exclusive rights. Ensure your agreement states that IBM cannot use or share your outputs or insights with other clients or for improving its services, except for your benefit alone.
Q: Does IBM indemnify AI-generated content?
A: It’s not automatic—this must be negotiated. IBM has begun offering IP indemnification for outputs of its own AI models; however, it is recommended to obtain this in writing. Push for an indemnity clause so IBM covers any IP infringement claims stemming from the AI’s output.
Q: How do I ensure data portability when leaving IBM Cloud?
A: Include exit provisions in the contract. Require IBM to provide all your data, conversation logs, and any custom model parameters in a usable format at termination. This ensures you can migrate your AI solution to another platform or in-house environment without losing work.
Read about SPSS licensing, IBM SPSS Licensing: Concurrent vs Named User and Subscription Options.
6. Five Recommendations for AI Contracting with IBM
- Negotiate Data Restrictions – Don’t Rely on Policy Defaults. Explicitly forbid IBM from using your data or prompts to train or enhance any AI beyond your own solution. This negotiation is crucial for safeguarding proprietary and sensitive information.
- Secure Model IP – Establish in writing that all AI outputs and custom models derived from your data are owned by your company. Clarify that IBM has no rights to these models or their insights, ensuring you retain exclusive control over the AI you build.
- Push for Indemnity – Demand that IBM assumes liability for IP risks from AI outputs. An indemnification clause will protect you if the AI inadvertently copies someone’s content. Use the fact that Microsoft, Google, and AWS offer similar protections as leverage in your discussions.
- Protect Residency & Portability – Insist on terms that ensure data residency (your data remains in specified regions/clouds) and portability (you can export models and data at any time). This shields you from compliance issues and prevents vendor lock-in by making sure you can move or remove your AI assets as needed.
- Use Leverage – Come to the table informed about competitor policies. If IBM is hesitant on a point (like indemnity or data use), cite the standards set by other providers and your willingness to choose a platform that offers the best terms. Leverage IBM’s desire to win or keep your business to negotiate the contract clauses that matter most to you.
By following these recommendations, you can confidently adopt IBM’s AI solutions, such as Watson and WatsonX, while minimizing risk.
Your data will remain yours, your AI models will be under your control, and IBM will stand behind its technology—leaving you free to innovate with peace of mind.
Read about our IBM Licensing Assessment Service.
