Fraud & impersonation
Recognising deception that learns.
How can defensive AI recognise deception when offensive AI generates and adapts fraudulent content?
A fraudulent interaction can combine persuasive messages, a claimed identity and a request that appears routine. This exploratory area asks how defensive AI can assess the whole interaction, explain suspicious signals and support proportionate verification. Controlled generative scenarios provide a way to test how detection changes when wording, context or impersonation methods change.
Research areas
Combine message content, identity evidence, interaction history and requested actions when assessing a possible fraud attempt.
Use controlled synthetic scenarios to test whether a defence recognises deception across different wording and contexts.
Investigate verification steps and human handoffs when the evidence is uncertain or the requested action is consequential.
Measure missed fraud, false alarms and the burden that defensive interventions place on legitimate users.
Publications & related resources
Projects, publications and research materials will be added here.
Contribute to this research.
Explore student research, internships and collaboration with the lab.