I’m a Computer Science Ph.D. student at the University of Wisconsin–Madison, working with Prof. Kassem Fawaz in the Wisconsin Privacy and Security Group.
My work sits between technical HCI, privacy, and AI-driven robotics. I’m interested in systems that more than one person uses, where a single set of privacy settings almost never fits everyone in the room. I lean on interviews and user studies about as much as on building things.
My dissertation asks one question: can GenAI understand a household’s privacy norms well enough to make decisions for the people living there? And if it can’t yet, what is missing?
Homes are a hard case. One account, one shared tablet, now a robot that moves from room to room, all serving people who want different things.
The work has gone in three directions so far.
I study how families actually use and mediate GenAI. That started with interviews of 12 households sharing a single ChatGPT login (Family Relations 2025, arXiv:2504.09004).
I test how well the models do on their own. Working from Contextual Integrity, 450 people answered scenarios about an in-home robot to set a human baseline, and ten LLMs then answered the same ones (arXiv:2507.16124). They are not ready to decide for anyone, though they improve a lot once a person is in the loop.
And I build what follows from that result. One detector read 12,962 public automation recipes and found 1,014 that could be turned into tools for surveillance or harassment (USENIX Security 2025). A robot that makes privacy calls on its own is the part I am still working toward.
I’ve given talks on this work at the University of Cambridge security seminar and, more recently, to threat assessment practitioners at the ATAP Great Lakes Chapter. Some of it has been picked up by WKOW, On Wisconsin, and the UW College of Engineering.
Email me at hzhang664@wisc.edu. My CV is here.