Roadmap for Mitigating Code Log Vulnerabilities via LLM-Driven Static Program Analysis

Roadmap for Mitigating Code Log Vulnerabilities via LLM-Driven Static Program Analysis. Krasniqi, R. & Shruti, A. C. In 2026 IEEE 36th International Symposium on Software Reliability Engineering Workshops (ISSREW), pages 1–4, Limassol, Cyprus, 2026. IEEE.
abstract bibtex

Traditional log security focuses on explicit token detection, such as scanning for leaked credentials. However, the most damaging vulnerabilities are inherently semantic. A code log statement may be syntactically flawless, yet dangerously misrepresent the program's true behavior. This semantic gap frequently arises when complex code dependencies and external program integrations obscure the program's underlying execution path, causing the code constructs embedded within the code logs to change semantically and rendering the code log's output unsafe. To mitigate these hidden flaws, this paper proposes a unified neuro-symbolic paradigm that interleaves the deterministic precision of static program analysis with the contextual reasoning of LLM models. We outline a three-tiered roadmap to achieve deep semantic alignment: (i) extracting AST-derived, log-centric code slices to capture precise structural dependencies; (ii) deploying fine-tuned, taxonomy-driven prompts to detect logical mismatches across distinct semantic vulnerability classes; and (iii) implementing an adversarial dual-model verification engine to autonomously cross-examine findings and suppress LLM hallucinations.

@inproceedings{krasniqi_roadmap_2026,
	address = {Limassol, Cyprus},
	title = {Roadmap for {Mitigating} {Code} {Log} {Vulnerabilities} via {LLM}-{Driven} {Static} {Program} {Analysis}},
	abstract = {Traditional log security focuses on explicit token detection, such as scanning for leaked credentials. However, the most damaging vulnerabilities are inherently semantic. A code log statement may be syntactically flawless, yet dangerously misrepresent the program's true behavior. This semantic gap frequently arises when complex code dependencies and external program integrations obscure the program's underlying execution path, causing the code constructs embedded within the code logs to change semantically and rendering the code log's output unsafe. To mitigate these hidden flaws, this paper proposes a unified neuro-symbolic paradigm that interleaves the deterministic precision of static program analysis with the contextual reasoning of LLM models. We outline a three-tiered roadmap to achieve deep semantic alignment: (i) extracting AST-derived, log-centric code slices to capture precise structural dependencies; (ii) deploying fine-tuned, taxonomy-driven prompts to detect logical mismatches across distinct semantic vulnerability classes; and (iii) implementing an adversarial dual-model verification engine to autonomously cross-examine findings and suppress LLM hallucinations.},
	language = {en},
	booktitle = {2026 {IEEE} 36th {International} {Symposium} on {Software} {Reliability} {Engineering} {Workshops} ({ISSREW})},
	publisher = {IEEE},
	author = {Krasniqi, Rrezarta and Shruti, Abanti Chakraborty},
	year = {2026},
	keywords = {Conference Workshop Papers},
	pages = {1--4},
}

Downloads: 0

{"_id":"t5zDqDa5yAZNZpAAT","bibbaseid":"krasniqi-shruti-roadmapformitigatingcodelogvulnerabilitiesviallmdrivenstaticprogramanalysis-2026","author_short":["Krasniqi, R.","Shruti, A. C."],"bibdata":{"bibtype":"inproceedings","type":"inproceedings","address":"Limassol, Cyprus","title":"Roadmap for Mitigating Code Log Vulnerabilities via LLM-Driven Static Program Analysis","abstract":"Traditional log security focuses on explicit token detection, such as scanning for leaked credentials. However, the most damaging vulnerabilities are inherently semantic. A code log statement may be syntactically flawless, yet dangerously misrepresent the program's true behavior. This semantic gap frequently arises when complex code dependencies and external program integrations obscure the program's underlying execution path, causing the code constructs embedded within the code logs to change semantically and rendering the code log's output unsafe. To mitigate these hidden flaws, this paper proposes a unified neuro-symbolic paradigm that interleaves the deterministic precision of static program analysis with the contextual reasoning of LLM models. We outline a three-tiered roadmap to achieve deep semantic alignment: (i) extracting AST-derived, log-centric code slices to capture precise structural dependencies; (ii) deploying fine-tuned, taxonomy-driven prompts to detect logical mismatches across distinct semantic vulnerability classes; and (iii) implementing an adversarial dual-model verification engine to autonomously cross-examine findings and suppress LLM hallucinations.","language":"en","booktitle":"2026 IEEE 36th International Symposium on Software Reliability Engineering Workshops (ISSREW)","publisher":"IEEE","author":[{"propositions":[],"lastnames":["Krasniqi"],"firstnames":["Rrezarta"],"suffixes":[]},{"propositions":[],"lastnames":["Shruti"],"firstnames":["Abanti","Chakraborty"],"suffixes":[]}],"year":"2026","keywords":"Conference Workshop Papers","pages":"1–4","bibtex":"@inproceedings{krasniqi_roadmap_2026,\n\taddress = {Limassol, Cyprus},\n\ttitle = {Roadmap for {Mitigating} {Code} {Log} {Vulnerabilities} via {LLM}-{Driven} {Static} {Program} {Analysis}},\n\tabstract = {Traditional log security focuses on explicit token detection, such as scanning for leaked credentials. However, the most damaging vulnerabilities are inherently semantic. A code log statement may be syntactically flawless, yet dangerously misrepresent the program's true behavior. This semantic gap frequently arises when complex code dependencies and external program integrations obscure the program's underlying execution path, causing the code constructs embedded within the code logs to change semantically and rendering the code log's output unsafe. To mitigate these hidden flaws, this paper proposes a unified neuro-symbolic paradigm that interleaves the deterministic precision of static program analysis with the contextual reasoning of LLM models. We outline a three-tiered roadmap to achieve deep semantic alignment: (i) extracting AST-derived, log-centric code slices to capture precise structural dependencies; (ii) deploying fine-tuned, taxonomy-driven prompts to detect logical mismatches across distinct semantic vulnerability classes; and (iii) implementing an adversarial dual-model verification engine to autonomously cross-examine findings and suppress LLM hallucinations.},\n\tlanguage = {en},\n\tbooktitle = {2026 {IEEE} 36th {International} {Symposium} on {Software} {Reliability} {Engineering} {Workshops} ({ISSREW})},\n\tpublisher = {IEEE},\n\tauthor = {Krasniqi, Rrezarta and Shruti, Abanti Chakraborty},\n\tyear = {2026},\n\tkeywords = {Conference Workshop Papers},\n\tpages = {1--4},\n}\n\n","author_short":["Krasniqi, R.","Shruti, A. C."],"key":"krasniqi_roadmap_2026","id":"krasniqi_roadmap_2026","bibbaseid":"krasniqi-shruti-roadmapformitigatingcodelogvulnerabilitiesviallmdrivenstaticprogramanalysis-2026","role":"author","urls":{},"keyword":["Conference Workshop Papers"],"metadata":{"authorlinks":{}}},"bibtype":"inproceedings","biburl":"https://api.zotero.org/users/10198036/collections/2RHJXKSI/items?key=X0RoN8iO9RtTbrWfSkRasb7b&format=bibtex&limit=100","dataSources":["37aX9ioouEvzbunGp","JHDShjsHrs6ZHE4bz"],"keywords":["conference workshop papers"],"search_terms":["roadmap","mitigating","code","log","vulnerabilities","via","llm","driven","static","program","analysis","krasniqi","shruti"],"title":"Roadmap for Mitigating Code Log Vulnerabilities via LLM-Driven Static Program Analysis","year":2026}