Revisiting Data Clauses to Protect AI-Generated Insights in Healthcare

Artificial intelligence (AI) has permeated nearly every aspect of commercial relationships in the healthcare industry. From drug discovery partnerships to clinical trial platforms, these collaborations increasingly rely on AI-enabled functionalities. However, many contracts governing these relationships were drafted before AI became a significant business concern. As a result, organizations may be inadvertently granting broader data rights than intended.

The Problem with Legacy Data Clauses

The issue lies not in whether a vendor, collaborator, or service provider can access the data-most agreements cover that. Instead, the critical question is what happens to the data after it's accessed:

• Can the recipient use the data to train AI models?
• Can they create derivative datasets?
• Who owns insights generated from these activities?
• Who owns the resulting outputs?

For many organizations, the answers are buried in contract provisions that received little scrutiny when the agreements were signed. Historically, data-use clauses focused on service performance, confidentiality obligations, and intellectual property ownership. AI changes this dynamic because data is no longer just an input necessary to fulfill a contract.

New Data Ownership Questions

Organizations are increasingly using datasets to train models, improve products, identify trends, develop new technologies, and generate commercially valuable outputs. Collaborations between companies often create new data that did not exist when the parties entered the relationship. This leads to a series of ownership questions that many legacy agreements do not adequately address:

• Who owns data generated through the collaboration?
  • New datasets created during the partnership may have different ownership rights than the original data.
• Who owns derivative datasets?
  • Datasets derived from the original data can be valuable and require clear ownership terms.
• May either party use the data to train AI models?
  • Training AI models with shared data can lead to intellectual property concerns.
• May AI-generated outputs be commercialized?
  • Outputs generated by AI, such as new drugs or treatments, need clear commercialization rights.
• Does the license survive termination?
  • Post-termination use of trained models and derived datasets should be explicitly defined.
• Can trained models continue to be used?
  • The ongoing use of trained models after a contract ends can have significant implications.

Navigating AI-Driven Security Threats

In addition to data ownership, healthcare organizations must also address the security risks associated with AI. Consultants Angela Burnett and Joel DuPlessis highlight how hospitals can navigate the rise of AI-driven security threats in health systems. These threats include:

• Data breaches: AI systems handling sensitive patient information are prime targets for cyberattacks.
• Model poisoning: Malicious actors can tamper with training data to influence AI model outcomes.
• Adversarial attacks: Attackers can manipulate input data to deceive AI models, leading to incorrect outputs.

To mitigate these risks, organizations should:

• Implement robust data encryption and access controls.
• Regularly audit AI systems for vulnerabilities.
• Train staff on best practices for handling sensitive data.

Key Takeaways

1. Revisit Data Clauses: Review existing contracts to ensure they address the new realities of AI data use.
2. Clarify Ownership Rights: Define who owns data generated through collaborations and derived datasets.
3. Address Commercialization: Specify whether AI-generated outputs can be commercialized and by whom.
4. Ensure Post-Termination Clarity: Determine the rights to trained models and derived datasets after a contract ends.
5. Enhance Security Measures: Implement robust security practices to protect against AI-driven threats.

By addressing these key points, healthcare organizations can better protect their data and maximize the value of their AI collaborations.