SunnyCONFERENCE TALKS
Research presented at IEEE Cyber-AI 2026. The conference papers and presentation certificates are linked below.
IEEE CYBER-AI 2026 · PRESENTED ONLINE
A Ranked Empirical Study of Twelve Comment Injection Attacks Against LLM-Based Code Vulnerability Detection
Ranks twelve attack comments in four families. The study finds that effectiveness follows the mechanism: comments claiming prior verification are especially effective.
IEEE CYBER-AI 2026 · PRESENTED ONLINE
Comment-Stripped Analysis Fully Defends LLM-Based Code Vulnerability Detectors Against Adversarial Comment Injection
Evaluates comment stripping as a practical defense. Across two open models and eight comment strategies, removing comments returns vulnerability detection to the clean-code rate without retraining.
Other accepted 2026 papers are listed with the full publication record on the existing Publications page. They are not listed here as presentations until a presentation is confirmed.
See all publications →