The cybersecurity landscape is rapidly evolving with the emergence of AI-powered malware. Recent research demonstrates that autonomous worms can leverage generative reasoning to dynamically adapt their attack strategies and self-replicate without human operators. Concurrently, Small Language Models (SLMs) are becoming highly capable, optimized specifically to operate locally on resource-constrained edge hardware.

The convergence of these trends introduces a severe new threat vector: SLM-powered malware targeting embedded Internet of Things (IoT) ecosystems. If adversaries deploy autonomous worms equipped with local SLMs, these malicious agents could independently reason through propagation decisions directly on the compromised host. This localized decision-making would allow the malware to spread autonomously across IoT networks while entirely eliminating the need for centralized Command and Control (C2) infrastructure, rendering many traditional network-based detection methods ineffective.

The task of the student is to develop a proof-of-concept worm capable of autonomously spreading from an embedded IoT device to another without the use of C2 infrastructure. All decisions of the worm has to be made by a small language model.

Tasks to be performed by the student will include:

- Present the state of the art on how latest malware use small and large language models
- Design an autonomously spreading worm that uses a small language model for making decisions on spreading and exploitation
- Implement the designed worm as a Proof-of-Concept
- Evaluate the implemented worm on small network of embedded IoT devices