User Privacy and LLMs

Currently, users do not have sufficient overview of what is publicly known about them on the Internet. They leave traces through their daily interactions with social media, merchandise sites, websites related to their profession or hobbies, and phone or address registries. We are designing and building a tool, called Online Privacy Pilot (OPP), that a user can use to collect all public information on the Internet that is related to herself. Thus, she will be ablet to gain an overview of what is known about herself and possibly select to remove some of this content from its source. The tool employs an LLM for helping the user decide which of the retrieved links are relevant to herself. Along the development of OPP, we also consider ethical considerations that accompany the use of such a tool: to collect information about the user herself and not about others. 

LLMs can offer users better control of their data (as described above), but they could also impose privacy risks. When a user employs a public LLM, the issued prompts might contain sensitive information. To prevent leaking this sensitive information to a third party that maintains such a public LLM, we are developing tools that automatically filter the prompts before they reach that LLM. At the same time, we study the impact that such filtering will have on usability.