PRI4GENAI

DESCRIPTION

This project adopts a human-centred approach to address user privacy in Generative AI (GenAI) systems, focusing specifically on those who use these technologies rather than those who develop them. The overarching goal is to identify, understand, and mitigate privacy risks faced by users when interacting with GenAI tools, particularly those based on text and image generation. The project considers users of different ages: children, teenagers, and adults.

This project is funded by the Ministerio de Ciencia, Innovación y Universidades through the program "Proyectos de Generación del Conocimiento 2023".

HOME

GOALS

The project aims to achieve the following specific objectives:

Assess the privacy risks faced by children, teenagers, and adults when using text and image GenAI.
Investigate the perceptions, experiences, and practices of these groups regarding privacy in GenAI.
Elicit the privacy norms that children, teenagers, and adults expect to be respected when using GenAI.
Design and develop a novel privacy assistant that learns these norms and helps users manage their privacy in text and image GenAI.
Create interventions (nudges) to raise awareness and help mitigate privacy risks in GenAI for users across all age groups.
Evaluate and refine both the privacy assistant and the interventions to ensure their effectiveness for children, teenagers, and adults.

Home

WORK TEAM

The work team includes the following academics from the Valencian Artificial Intelligence Research Institute (VRAIN) at the Universitat Politècnica de València (UPV): Jose Such, Elena del Val, Caterina Maidhof, Gonzalo Méndez, Vicente Botti, Ana García-Fornés, Estefanía Argente, Soledad Valero, Emilio Vivancos, Agustín Espinosa and Vicente Cabedo.

Home

MEDIA

Un equipo del Instituto VRAIN trabaja para proteger la privacidad de las personas en aplicaciones de IA generativa. Link to the news

¿Conoces el proyecto PRI4GENAI?Link to the news

PUBLICATIONS

Xiao Zhan, Juan-Carlos Carrillo, William Seymour, Jose Such. Malicious LLM-Based Conversational AI Makes Users Reveal Personal Information. En 34th USENIX Security Symposium. USENIX Association, 2025.

Caterina Maidhof, Elena del Val, Jose Such: Privacy Risks for Underaged User Groups in LLM-based Conversational Agent Interactions. GenAICHI, 2025.

HOME

EDUCATIONAL WORSHOPS

We are developing educational materials to better understand how generative AI chatbots work and what aspects need to be considered to protect our privacy when interacting with them.