Privacy-First AI

AI assistants are most useful when they know something about your work, and that's exactly what makes them risky. Whether you're typing a prompt, uploading a document, connecting a data source, or letting an AI agent work across your files, every interaction is a disclosure decision: student records, unpublished research, personnel matters, and confidential institutional data don't stop being sensitive just because they've entered an AI tool. This workshop builds the habit of thinking before you share, treating everything an AI system can see as a deliberate choice rather than an afterthought. No technical background required. You'll learn to weigh what information an AI tool actually needs against the risk of sharing it - using Duke's data classification policy as a guide, alongside your own judgment about context, sensitivity, and consequences. You'll also configure the privacy settings that matter in the tools you already use, and practice concrete techniques for getting high-quality AI output from de-identified, minimized, or abstracted inputs. You'll leave with a practical decision framework for what to share and what to hold back, hands-on experience applying it to real-world scenarios, and the confidence to make AI a default part of your work while keeping sensitive information exactly where it belongs.