How far have we been on the path of LLM-enhanced vulnerability detection
Open accessThe DOI matches its Crossref record; 11 of the 11 references it cites were found in a registry or an index.
Abstract
Large Language Models(LLMs) have been proposed and deployed for several years. With the rapid development of LLMs, security researchers explores its effectiveness in vulnerability detection. In this paper, we summarize all the papers involving LLM-enhanced vulnerability detection crossing Top-tier software engineering/security conferences/journals, such as USENIX Security, NDSS, ACM CCS and IEEE S&P from security areas, ICSE, FSE and ASE from software engineering, which include nearly 50 papers to discuss how far have we been on the path of LLM-enhanced vulnerability detection. Besides each work’s detailed content, we also discuss the limitations they suffer and the promising future research direction.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“Of the thirteen agentic vulnerability-discovery studies coded in our corpus (eight single-agent and five multi-agent; Table 5; non-agentic baselines and feasibility/tool-only entries in the coded matrix are outside this count), only two report any quantitative evidence about findings that fail to reproduce under independent verification [123,124]; the remaining eleven report detection-style”
From Agentic AI and Large Language Models for Autonomous IoT Cybersecurity: A Systematic Survey, Taxonomy, and Research Roadmap (2026) · cites this paper as [123]
In the citing paper
“Multi-agent smart-contract auditors [119,121-123,125] now match or exceed expert auditors on standard benchmarks.”
From Agentic AI and Large Language Models for Autonomous IoT Cybersecurity: A Systematic Survey, Taxonomy, and Research Roadmap (2026) · cites this paper as [123]
Citations
Cites
API-misuse detection driven by fine-grained API-constraint knowledge graph
Ren, X., Ye, X., Xing, Z., Xia, X., Xu, X., Zhu, L., & Sun, J. (2020) · IEEE/ACM International Conference on Automated Software Engineering (ASE) · doi:10.1145/3324884.3416551 · Cited by 90 · cited 1×
“For static detection, researchers are encouraged to detect bugs before running programs by analyzing documents [14, 16], source code [8, 10] or patches [17].”
When Fuzzing Meets LLMs: Challenges and Opportunities
Jiang, Y., Liang, J., Ma, F., Chen, Y., Zhou, C., Shen, Y., Wu, Z., Fu, J., Wang, M., Li, S., et al. (2024) · ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering (ESEC/FSE) · doi:10.1145/3663529.3663784 · Cited by 75
APEx: automated inference of error specifications for C APIs
Kang, Y., Ray, B., & Jana, S. (2016) · IEEE/ACM International Conference on Automated Software Engineering (ASE) · doi:10.1145/2970276.2970354 · Cited by 49 · cited 1×
“For static detection, researchers are encouraged to detect bugs before running programs by analyzing documents [14, 16], source code [8, 10] or patches [17].”
Lv, T., Li, R., Yang, Y., Chen, K., Liao, X., Wang, X., Hu, P., & Xing, L. (2020) · ACM SIGSAC Conference on Computer and Communications Security (CCS) · doi:10.1145/3372297.3423360 · Cited by 43 · cited 1×
“For static detection, researchers are encouraged to detect bugs before running programs by analyzing documents [14, 16], source code [8, 10] or patches [17].”
Self-citationLLM-Powered Static Binary Taint Analysis
Liu, P., Sun, C., Zheng, Y., Feng, X., Qin, C., Wang, Y., Xu, Z., Li, Z., Di, P., Jiang, Y., et al. (2025) · ACM Transactions on Software Engineering and Methodology · doi:10.1145/3711816 · Cited by 39
Self-citationWu, Y., Wen, M., Yu, Z., Guo, X., & Jin, H. (2024) · IEEE/ACM International Conference on Automated Software Engineering (ASE) · doi:10.1145/3691620.3695013 · Cited by 34
The Digital Cybersecurity Expert: How Far Have We Come?
Wang, D., Zhou, G., Li, X., Bai, Y., Chen, L., Qin, T., Sun, J. Q., & Li, D. (2025) · IEEE Symposium on Security and Privacy (S&P) · doi:10.1109/sp61157.2025.00198 · Cited by 17
RecurScan: Detecting Recurring Vulnerabilities in PHP Web Applications
Shi, Y., Zhang, Y., Bai, T., Zhang, L., Tan, X., & Yang, M. (2024) · International World Wide Web Conference (WWW) · doi:10.1145/3589334.3645530 · Cited by 16 · cited 1×
“For static detection, researchers are encouraged to detect bugs before running programs by analyzing documents [14, 16], source code [8, 10] or patches [17].”
Lin, L. G., Zhu, Q., Chen, H., Wang, Z., Wu, R., & Xie, X. (2025) · Proceedings of the ACM on software engineering. · doi:10.1145/3728908 · Cited by 11 · cited 1×
“Researcher [12] introduces QTRAN, a novel approach that leverages LLMs to automatically extend Metamorphic Testing, across diverse platforms.”
Safe4U: Identifying Unsound Safe Encapsulations of Unsafe Calls in Rust using LLMs
Li, H., Wang, B., Hu, X., & Xia, X. (2025) · Proceedings of the ACM on software engineering. · doi:10.1145/3728890 · Cited by 7 · cited 1×
“Recent work [11] presents Safe4U, a novel framework designed to identify unsound safe encapsulations of unsafe calls (EUCs) in Rust code.”
Research works printed but not linked: 2
- Lezhi Ma Shangqing Liu Yi Li Xiaofei Xie and Lei Bu. 2024. Specgen: Automated generation of formal program specifications via large language models. arXiv preprint arXiv:https://arXiv.org/abs/2401.08807 (2024). · arXiv
- Ashish Vaswani Noam Shazeer Niki Parmar Jakob Uszkoreit Llion Jones Aidan N Gomez Łukasz Kaiser and Illia Polosukhin. 2017. Attention is all you need. Advances in neural information processing systems 30 (2017).
Cited by
Nageshwaran, V., & Ezekiel, S. (2026) · Electronics · doi:10.3390/electronics15122740 · Cited by 4
“Multi-agent smart-contract auditors [119,121–123 , Compared to coverage-guided fuzzers such as AFL++ and libFuzzer—which excel at exploring branch space within a single binary but treat every input byte as opaque—LLM-driven discovery contributes three capabilities.”
InfluentialNageshwaran, V., & Ezekiel, S. (2026) · doi:10.20944/preprints202606.0142.v1 · Cited by 4
Evaluating Large Language Models in Cybersecurity: A Systematic Taxonomy and Empirical Analysis
Chen, M., Cheng, H., Su, T., Chen, M., Cai, W., & Zou, H. (2026) · Electronics · doi:10.3390/electronics15102222 · Cited by 2
“Initial studies primarily assessed foundational vulnerability detection under zero-shot conditions [3,29,30], while concurrent research identified endogenous risks in AI coding assistants [31,32].”
Metadata
- Authors
- Shen, W., Yang, Y., Zhang, X., Cui, L., & Wang, Y.
- Affiliations
- Guangdong Power Grid Company (China) (CN)
- Chinese Academy of Sciences (CN)
- Venue
- Proceedings of the 2025 International Symposium on Artificial Intelligence and Computational Social Sciences
- Publisher
- ACM
- Published
- 2025-09-19
- Pages
- 651-657
- OpenAlex
- W4417504582
- Cited by
- 2
- References
- 11
- Citations by year
- 2026: 2
- FWCI
- 1.56
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- Locations
- International Symposium on Artificial Intelligence and Computational Social Sciences (AICSS) (publishedVersion, cc-by)
- Topics
- Physical Sciences › Computer Science › Software › Software Testing and Debugging Techniques
- Physical Sciences › Computer Science › Information Systems › Web Application Security Vulnerabilities
- Physical Sciences › Computer Science › Information Systems › Software Engineering Research
- Keywords
- large language models, vulnerability detection, software security, research survey, research limitations, future research directions
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: How far have we been on the path of LLM-enhanced vulnerability detection. https://app.cochat.ai/papers/10.1145/3776759.3776861
- The paper
- Shen, W., Yang, Y., Zhang, X., Cui, L., & Wang, Y. (2025). How far have we been on the path of LLM-enhanced vulnerability detection. Proceedings of the 2025 International Symposium on Artificial Intelligence and Computational Social Sciences. https://doi.org/10.1145/3776759.3776861
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.