Trial lecture
Prompt injection attacks: mechanisms, risks and defences
Time and place: June 15, 2026 11:15 AM. Kristen Nygaards sal (5370), Ole-Johan Dahls hus
Main research findings
Modern software is no longer just code and databases; AI now sits at the core of applications across industries. This shift introduces a new class of security vulnerabilities that traditional approaches were not designed to address. This thesis proposes SAIL-K, a defense-in-depth framework that secures AI-powered applications across five layers: at the Social layer, how people interact with AI; at the Application Infrastructure layer, how the threat landscape is compounded by cloud environments; at the Backend Logic and Database layer, how application logic can be targeted; at the AI Model layer, exploitable failure modes in reasoning; and at the AI Knowledge layer, problematic knowledge that AI may learn from and its implications for end users. Together, these layered mitigations compose a more resilient deployment posture for modern AI systems.
Adjudication committee
- Professor Michael Lones, Heriot-Watt University, UK
- Dr. Siri Bromander, Mnemonic, Norway
- Professor Stephan Oepen, University of Oslo, Norway
Supervisors
Associate Professor Åvald Åslaugson Sommervoll, University of Oslo, Norway
Associate professor Fabio Massimo Zennaro, University of Bergen, Norway
Associate Professor László Erdődi, University of Oslo, Norway
Chair of defence
- Professor Nils Gruschka, Department of Informatics, University of Oslo
Contact information at Department: Mozhdeh Sheibani Harat