Program Analysis Guided LLM Agent for Proof-of-Concept Generation

Achintya Desai, Md Shafiuzzaman, Wenbo Guo, Tevfik Bultan · 2026

Read Program Analysis Guided LLM Agent for Proof-of-Concept Generation by Achintya Desai, Md Shafiuzzaman, Wenbo Guo, Tevfik Bultan (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-07624.txt.

Description

Desai and three colleagues present PAGENT, a scalable approach to generating proof-of-concept inputs that trigger reported software vulnerabilities. Given project source code and a suspect code location, PAGENT blends lightweight rule-based static analysis with sanitizer-based profiling and coverage data to guide a generation agent. Symbolic execution and fuzzing need expert guidance and manual steps and scale poorly, while plain language-model generation succeeds rarely. In experiments, the hybrid approach beats the prior top agentic approach by 132 percent on proof-of-concept generation.

Catalog record

CREATORAchintya Desai, Md Shafiuzzaman, Wenbo Guo, Tevfik Bultan
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE64491
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.07624

Browse the full library · About the Tower of Babel