Working with Sensitive Data in Secure GPT: What You Can and Can't Do

Tool used: Stanford Secure GPT

Secure GPT is approved for high-risk data, but approval to use a tool isn't the same as knowing how to use it well. In this 90-minute session, you'll explore the boundaries of Secure GPT's appropriate use, examine how data classification and institutional policies shape your workflows, and work through realistic scenarios involving PII, PHI, and other sensitive information. Together, we'll examine practical applications for research, administration, and reporting while building confidence in responsible AI use.

Learners will have the opportunity to:

1. Explore Secure GPT's capabilities and approved use cases for sensitive data
2. Practice classifying data and matching it to the appropriate Stanford AI tool
3. Experiment with prompting strategies that protect data integrity and produce reliable outputs
4. Work through realistic scenarios involving PII, PHI, and other protected information
5. Consider downstream implications for sharing, publishing, and reporting on sensitive data

- Stanford data risk classifications and where Secure GPT fits
- What Secure GPT is approved for, and what it isn't
- Prompting strategies for working with sensitive records
- Distinguishing inputs you can use from outputs you can share
- Recognizing re-identification risk in AI-generated content
- Practical scenarios across research, administration, and reporting
- Documentation, audit trails, and responsible AI practices