Malagnino, Carlo (A.A. 2024/2025) PII sanitization in LLMs: leveraging format-preserving encryption. Tesi di Laurea in Machine learning, Luiss Guido Carli, relatore Giuseppe Francesco Italiano, pp. 52. [Master's Degree Thesis]

PDF (Full text) Restricted to Registered users only Download (3MB) | Request a copy

Abstract/Index

LLMs overview and applications. Introduction to LLMs. Applications of LLMs across industries. Data sensitivity and privacy risks in LLM workflows. Limitations of current solutions. FPE for PII sanitization. Introduction to personal identifiable information (PII). Definition and characteristics of FPE. Comparison with traditional encryption methods. Vaultless shield framework. Framework overview. Architectural components. Data flow across the LLM lifecycle. Contextual understanding retention in prompting. Semi-stateless design and database support. Security analysis of the framework. Membership inference attacks. Model inversion attacks. Data extraction attacks. Proposed framework effectiveness analysis. Regulatory compliance and legal considerations. EU regulatory landscape. Alignment with GDPR principles.

References

Bibliografia: pp. 50-52.

Thesis Type:	Master's Degree Thesis
Institution:	Luiss Guido Carli
Degree Program:	Master's Degree Programs > Master's Degree Program in Data Science e Management (LM-91)
Chair:	Machine learning
Thesis Supervisor:	Italiano, Giuseppe Francesco
Thesis Co-Supervisor:	Di Somma, Simone
Academic Year:	2024/2025
Session:	Summer
Deposited by:	Alessandro Perfetti
Date Deposited:	11 Sep 2025 13:52
Last Modified:	11 Sep 2025 13:52
URI:	https://tesi.luiss.it/id/eprint/43150

Downloads

Repository Staff Only

View Item