My research focuses on making AI-assisted software development more trustworthy, with particular attention to the quality, security, robustness, and reliability of AI-generated code.

AI-driven exploit generation. I investigate how generative models can support offensive security by producing exploits from natural-language descriptions. This includes designing automated frameworks to assess their syntactic and semantic correctness and exploring context-aware generation approaches.

Security and robustness of code generators. I study how prompt ambiguity, natural-language perturbations, and poisoned training data affect code generation systems. My work also explores data-augmentation, detection, and mitigation strategies that make these models more resilient.

Code quality assessment and enhancement. I examine how training-data quality influences the correctness, security, and maintainability of generated software. I use static analysis and large-scale empirical comparisons between human-written and AI-generated code to identify defects and vulnerabilities and inform better data-curation practices.

Across these directions, my goal is to develop dependable AI-assisted software engineering techniques that produce functionally correct, secure, and maintainable code.

You can reach out at cristina.improta@unina.it or find me at DESSERT lab, Via Claudio 21, Naples, IT, Building 3/A, 4th floor.