Alessio Giummarra
Contextual Data Loss Prevention: An LLM-Driven Approach.
Rel. Franco Galante, Matteo Boffa. Politecnico di Torino, Corso di laurea magistrale in Cybersecurity, 2026
Abstract
Data Loss Prevention (DLP) tools have the objective of preventing employees from sending sensitive information outside their company. The importance of this task is evident: a single leaked message can expose an unannounced acquisition, a customer list, or financial results before they are public. However, it is also hard to get right: a tool that blocks too much buries people in false alarms, and they stop trusting it. Traditionally, DLP systems analyse the syntactic elements of the text. A rule is able to identify an IBAN, a credit card number, or an email address, because these follow a fixed pattern. What the text is actually about is not taken into account.
An employee can describe an acquisition that has not been announced yet without writing a single string that a rule would catch
Relatori
Anno Accademico
Tipo di pubblicazione
Numero di pagine
Informazioni aggiuntive
Corso di laurea
Classe di laurea
URI
![]() |
Modifica (riservato agli operatori) |
