Background

Recent advances in information and communication technology have increased the complexity and connectivity of modern software systems. As Artificial Intelligence (AI) becomes more widely adopted, more sophisticated AI-based attacks (i.e., offensive AI) on systems are emerging. These attacks can be adaptive, harder to detect, and highly destructive, and existing defence mechanisms are often inadequate against their evolving nature and decision logic. Systems that rely on interconnected components, external services, and third-party technologies further expand the attack surface, making it critical for organisations to prepare and strengthen resilience against offensive AI.

Description

This thesis contributes to resilience against AI-based cyberattacks. The overall goal is to understand which types of AI-based cyberattacks may affect software systems, their potential impact, and the strategies that can protect systems from such attacks. The proposed approach might employ an AI attacker and an AI defender that engage in a game-theoretic setting: both agents act on a target system, receive rewards through reinforcement learning, and adapt their strategies over time. The AI attacker aims to uncover vulnerabilities and improve attack strategies, while the AI defender learns to adapt responses and strengthen system resilience against offensive AI.

Key Responsibilities

In this thesis, the student will review existing work on offensive AI and AI-based cybersecurity defence, with an emphasis on game theory and reinforcement learning. The student will then design and prototype an approach in which an AI attacker and an AI defender interact with a software system. The focus will be on developing methods to:

• model the interaction between an AI attacker and an AI defender using game-theoretic concepts,

• apply reinforcement learning so that both agents improve their strategies based on actions, rewards, and the target system’s state, and

• evaluate the approach with respect to attack effectiveness, defensive adaptation, and overall system resilience.

The expected outcome is a prototype framework that demonstrates how offensive and defensive AI can co-evolve and provides insights into strategies for protecting software systems against AI-based cybersecurity attacks.

Qualifications

Candidates are expected to be enrolled in a master's programme in a field related to computer science and engineering at a Swedish university. Having already completed AI-related courses and/or gained work experience with AI/ML (e.g., reinforcement learning) is an advantage. Knowledge of cybersecurity is also an advantage.

Conditions

• Location: RISE, Gothenburg

• Some physical presence is expected.

• Applications are reviewed on a rolling basis; apply as soon as possible, but no later than October 31st, 2026.

• Starting date: January 2027.

• Credits: 30 ECTS

• Compensation: 10.000 SEK upon successful completion of the thesis.

Welcome with your application!

Contact: Dr. Rodi Jolak +46 10 228 42 56