Potential Misuse and Data Sharing

Generative AI has potential applications across medical school, health professions education, health care and beyond. However, there are also concerns about the potential misuse of these tools and data shared with the services. When you provide information to these tools, such as queries, student essays, grant proposals, source code or datasets, it is the same as posting information on a public website.

WVSOM encourages the use of generative AI services, as long as no institutional data is submitted to them without approval. Currently, Microsoft 365 Copilot is the only generative AI permitted. Any other system used for official institutional data or information will need to go through review to ensure the necessary contracts and safeguards are in place to protect the data submitted and to ensure the algorithms in use are ethical, transparent and beneficial to the WVSOM community.

Types of institutional or patient/clinical data that should not be submitted to public versions of generative AI tools, even when anonymized, include:

• Data classified as Institutional Data or higher (for examples, visit the Procedure For Access to Institutional Data)
• Data protected under HIPAA (for example, patient data; refer to the WVSOM HIPAA policy for details)
• Data that may be considered student, faculty or staff intellectual property, unless the individual submitting that intellectual property created it

Specific examples that are not appropriate for the public versions of generative AI tools include:

• Sharing names and information about a real student, employee, research participant or patient
• Asking an unsupported/approved AI model to summarize and grade a student paper or assignment
• Sharing employee-related data such as performance or benefit information for communication drafting or analysis
• Asking an unsupported AI model to generate code for WVSOM systems protecting institutional data or sharing WVSOM source code for editing
• Sharing grant proposals still under review