Research

Research in secure software engineering, LLM security, and AI/ML security.

LLMs for security

LLM4SEC

03

Secure Code Generation

We study how language models can generate code that satisfies both functional requirements and security constraints. Validation is used to identify weaknesses early.

  • Security-Aware Generation
  • Functional Correctness
  • Code Validation

Current work Secure code generation under functional and security requirements

04

Kernel Code Fault Localization

We are building a benchmark for locating faults in kernel code, where low-level behaviour and long-range dependencies challenge conventional methods.

  • Linux Kernel
  • Fault Localization
  • Benchmarking

Current work LLM-based fault localization for Linux kernel code

Security for LLMs and AI/ML systems

SEC4LLM & AI/ML Security

01

RAG Security

We examine attacks on retrieval-augmented generation, including malicious or misleading retrieved content. Attacks and defences are compared under shared evaluation criteria.

  • Retrieval Poisoning
  • Prompt Injection
  • Attack–Defence Evaluation

Related paper Security Analysis and Defense Comparison in RAG

02

IaC Agent Assurance

We evaluate AI-generated infrastructure-as-code beyond syntax and execution, asking whether it matches user intent and aligns with relevant security policies.

  • Terraform
  • Intent Alignment
  • Security Policy Compliance

Current work Intent- and policy-aware evaluation of Terraform-generating agents