← Back to Projects

LLM Comment Vulnerability Study

Collaborative AI security research exploring how misleading code comments can influence LLM outputs. The project includes a public research site, dataset resources, and documented findings, and was accepted and presented.

Role
Research Contributor
Team
Research team collaboration
Duration
Completed
Company
Research Project

My Contribution

  • Contributed to research design and methodology for LLM comment-based vulnerabilities
  • Supported dataset preparation and analysis for adversarial prompt testing
  • Built and maintained the public research site experience
  • Collaborated on documentation and presentation materials

Key Highlights

  • Public research site summarizing methodology, datasets, and results
  • Dataset resources hosted with citations for reuse
  • Accepted and presented research outcomes

Impact

  • Advanced awareness of LLM safety risks from misleading code comments
  • Provided reusable dataset resources for the AI safety community
  • Demonstrated collaborative research and delivery to publication-quality output

Tech Stack / Tools

PythonMachine LearningPyTorchOpenAI APISecurity TestingResearch Methodology