{"slug": "when-llm-decompilers-recompile-more-and-preserve-less", "title": "When LLM Decompilers Recompile More and Preserve Less", "summary": "A 4 Sep 2026 arXiv paper, \"When LLM Decompilers Recompile More and Preserve Less,\" reports that LLM-based decompilers can pass every shipped recompilability and re-executability test while still diverging behaviorally from the original binary. The authors' Decompile-Diverge oracle, which synthesizes a driver per function and fuzzes a corpus from the reference, found candidates passing all shipped tests still diverge on 4.9% of inputs overall and up to 13% for a single system across eight systems in nine configurations; on 300 real GitHub library functions the strongest refinement LLM raised Ghidra's build rate from 75% to 90% while its Matched rate fell from 74% to 62%, and on 287 CVE-grounded functions up to one tenth showed Crash Absence. The paper attributes the divergence to introduced fields, types, callees, and guards that replace the visible unknowns traditional decompilers like Ghidra and Hex-Rays leave as placeholders.", "body_md": "# Computer Science > Cryptography and Security\n\n  [Submitted on 4 Sep 2026]\n\n# Title:When LLM Decompilers Recompile More and Preserve Less\n\n[View PDF](https://arxiv.org/pdf/2609.05370)\n\n[HTML (experimental)](https://arxiv.org/html/2609.05370v1)\n\nAbstract:Decompilation recovers high-level source from compiled machine code and serves as a foundation for security tasks such as vulnerability detection and malware analysis. Traditional decompilers like Ghidra and Hex-Rays expose whatever they cannot resolve as visible placeholders and often emit pseudocode that will not compile or execute; LLM-based decompilers produce clean, idiomatic C and are now judged almost entirely by recompilability and re-executability: whether the output builds and passes its shipped input/output tests. We show that these metrics can reward the wrong path: a function may recompile and pass every shipped test yet diverge on other legitimate inputs, and a disclosed vulnerability may disappear from the recompiled code with no visible trace of the crash. Neither failure is caught by existing suites.\n\nTo address this gap, we propose Decompile-Diverge, a behavioral comparison oracle not relying on fixed or hand-crafted tests: for each function it synthesizes a driver, grows a fuzzing corpus from the reference, and reruns the decompiled code on the same inputs to detect changes in the function's behavior. Across eight systems in nine configurations on established LLM decompilation corpora, candidates that pass every shipped test still diverge from the original on our input corpus: 4.9% overall, and as many as 13% for a single system. On 300 real GitHub library functions and 287 CVE-grounded functions, recompilability and behavioral agreement can come apart: the strongest refinement LLM lifts Ghidra's build rate from 75% to 90%, while its Matched rate falls from 74% to 62%; on disclosed vulnerabilities, up to one tenth exhibit Crash Absence in its output. Source-level analysis traces this divergence to introduced fields, types, callees, and guards that replace the visible unknowns traditional tools leave behind.\n\n### References & Citations\n\nLoading...\n\n# Bibliographic and Citation Tools\n\nBibliographic Explorer \n\n*(*[What is the Explorer?](https://info.arxiv.org/labs/showcase.html#arxiv-bibliographic-explorer))\nConnected Papers \n\n*(*[What is Connected Papers?](https://www.connectedpapers.com/about))\nLitmaps \n\n*(*[What is Litmaps?](https://www.litmaps.co/))\nscite Smart Citations \n\n*(*[What are Smart Citations?](https://www.scite.ai/))\n# Code, Data and Media Associated with this Article\n\nalphaXiv \n\n*(*[What is alphaXiv?](https://alphaxiv.org/))\nCatalyzeX Code Finder for Papers \n\n*(*[What is CatalyzeX?](https://www.catalyzex.com))\nDagsHub \n\n*(*[What is DagsHub?](https://dagshub.com/))\nGotit.pub \n\n*(*[What is GotitPub?](http://gotit.pub/faq))\nHugging Face \n\n*(*[What is Huggingface?](https://huggingface.co/huggingface))\nScienceCast \n\n*(*[What is ScienceCast?](https://sciencecast.org/welcome))\n# Demos\n\n# Recommenders and Search Tools\n\nInfluence Flower \n\n*(*[What are Influence Flowers?](https://influencemap.cmlab.dev/))\nCORE Recommender \n\n*(*[What is CORE?](https://core.ac.uk/services/recommender))\n# arXivLabs: experimental projects with community collaborators\n\narXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.\n\nBoth individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.\n\nHave an idea for a project that will add value for arXiv's community? [**Learn more about arXivLabs**](https://info.arxiv.org/labs/index.html).", "url": "https://wpnews.pro/news/when-llm-decompilers-recompile-more-and-preserve-less", "canonical_source": "https://arxiv.org/abs/2609.05370", "published_at": "2026-10-12 02:44:55+00:00", "updated_at": "2026-10-12 03:06:42.279294+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-research", "ai-safety", "machine-learning"], "entities": ["Decompile-Diverge", "Ghidra", "Hex-Rays", "arXiv", "GitHub", "CVE"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/when-llm-decompilers-recompile-more-and-preserve-less", "markdown": "https://wpnews.pro/news/when-llm-decompilers-recompile-more-and-preserve-less.md", "text": "https://wpnews.pro/news/when-llm-decompilers-recompile-more-and-preserve-less.txt", "jsonld": "https://wpnews.pro/news/when-llm-decompilers-recompile-more-and-preserve-less.jsonld"}}