{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,7,13]],"date-time":"2026-07-13T13:18:09Z","timestamp":1783948689651,"version":"3.55.0"},"reference-count":0,"publisher":"Association for the Advancement of Artificial Intelligence (AAAI)","issue":"1","content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["AIES"],"abstract":"<jats:p>Automated detection of vulnerabilities in source code is an\nessential cybersecurity challenge, underpinning trust in\ndigital systems and services. Graph Neural Networks (GNNs)\nhave emerged as a promising approach as they can learn the\nstructural and logical code relationships in a data-driven\nmanner. However, the performance of GNNs is severely\nlimited by training data imbalances and label noise. GNNs\ncan often learn \u201cspurious\u201d correlations due to superficial\ncode similarities in the training data, leading to\ndetectors that do not generalize well to unseen real-world\ndata. In this work, we propose a new unified framework for\nrobust and interpretable vulnerability detection\u2014that we\ncall VISION\u2014to mitigate spurious correlations by\nsystematically augmenting a counterfactual training\ndataset. Counterfactuals are samples with minimal semantic\nmodifications that have opposite prediction labels. Our\ncomplete framework includes: (i) generating effective\ncounterfactuals by prompting a Large Language Model (LLM);\n(ii) targeted GNN model training on synthetically paired\ncode examples with opposite labels; and (iii) graph-based\ninterpretability to identify the truly crucial code\nstatements relevant for vulnerability predictions while\nignoring the spurious ones. We find that our framework\nreduces spurious learning and enables more robust and\ngeneralizable vulnerability detection, as demonstrated by\nimprovements in overall accuracy (from 51.8% to 97.8%),\npairwise contrast accuracy (from 4.5% to 95.8%), and\nworst-group accuracy increasing (from 0.7% to 85.5%) on the\nwidely popular Common Weakness Enumeration (CWE)-20\nvulnerability. We also demonstrate improvements using our\nproposed metrics, namely, intra-class attribution variance,\ninter-class attribution distance, and node score\ndependency. We provide a new benchmark for vulnerability\ndetection, CWE-20-CFA, comprising 27,556 samples from\nfunctions affected by the high-impact and frequently\noccurring CWE-20 vulnerability, including both real and\ncounterfactual examples. Furthermore, our approach enhances\nsocietal objectives of transparent and trustworthy AI-based\ncybersecurity systems through interactive visualization for\nhuman-in-the-loop analysis.<\/jats:p>","DOI":"10.1609\/aies.v8i1.36592","type":"journal-article","created":{"date-parts":[[2025,10,15]],"date-time":"2025-10-15T13:18:33Z","timestamp":1760534313000},"page":"812-823","source":"Crossref","is-referenced-by-count":4,"title":["VISION: Robust and Interpretable Code Vulnerability Detection Leveraging Counterfactual Augmentation"],"prefix":"10.1609","volume":"8","author":[{"given":"David","family":"Egea","sequence":"first","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Barproda","family":"Halder","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Sanghamitra","family":"Dutta","sequence":"additional","affiliation":[],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"9382","published-online":{"date-parts":[[2025,10,15]]},"container-title":["Proceedings of the AAAI\/ACM Conference on AI, Ethics, and Society"],"original-title":[],"link":[{"URL":"https:\/\/ojs.aaai.org\/index.php\/AIES\/article\/download\/36592\/38730","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/ojs.aaai.org\/index.php\/AIES\/article\/download\/36592\/38730","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2025,10,15]],"date-time":"2025-10-15T13:18:34Z","timestamp":1760534314000},"score":1,"resource":{"primary":{"URL":"https:\/\/ojs.aaai.org\/index.php\/AIES\/article\/view\/36592"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025,10,15]]},"references-count":0,"journal-issue":{"issue":"1","published-online":{"date-parts":[[2025,10,15]]}},"URL":"https:\/\/doi.org\/10.1609\/aies.v8i1.36592","relation":{},"ISSN":["3065-8365"],"issn-type":[{"value":"3065-8365","type":"electronic"}],"subject":[],"published":{"date-parts":[[2025,10,15]]}}}