{"id":8855,"date":"2026-11-13T08:15:48","date_gmt":"2026-11-13T08:15:48","guid":{"rendered":"https:\/\/i10xnews.kinsta.cloud\/small-decision-model-memory-safety-bugs-study"},"modified":"2026-11-13T08:15:48","modified_gmt":"2026-11-13T08:15:48","slug":"small-decision-model-memory-safety-bugs-study","status":"publish","type":"post","link":"https:\/\/i10x.ai\/news\/small-decision-model-memory-safety-bugs-study","title":{"rendered":"Study: A Small Decision Model Helped Find 187 Memory-Safety Bugs"},"content":{"rendered":"\n<h2 class=\"wp-block-heading\">\u26a1 Quick Take<\/h2>\n\n\n<ul class=\"wp-block-list\"><li><strong>The finding:<\/strong> Intern-Decision-4B, a 4-billion-parameter open-weight decision model, ranked candidate functions in a deterministic search harness and helped a campaign confirm 187 memory-safety findings across more than 4,000 C and C++ projects. By worst-case CVSS v3.1, 12 are critical, 148 high, 23 medium and 4 low. The search used no frontier model.<\/li><li><strong>Why it matters:<\/strong> Alone, the model is weak: it separates vulnerable from benign functions with AUC 0.73, but a vulnerable function from its own patch with only 0.56. Its value is in ordering queues. On 75 public CVEs from projects without any of the campaign&#8217;s findings, the fixed function reached its repository&#8217;s top 25 in 21% of cases, 1.4 times as often as random order of the same queues.<\/li><li><strong>What it means:<\/strong> For this model and harness, a small decision model supports vulnerability discovery through prioritization, while deterministic analysis and verification remain essential. All confirmed findings are being disclosed to maintainers under coordinated disclosure.<\/li><\/ul>\n\n\n<h2 class=\"wp-block-heading\">In plain English<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">A team at Cyrex and Deus Rex asked whether a small, open AI model, one that only scores a short list of answers instead of writing text, can help find memory bugs in real C and C++ software. On its own the model is weak: it barely tells a buggy function from the patched version of the same function. Used to rank a list of already flagged functions, it helped reviewers find more bugs near the top of the list. A campaign on more than 4,000 projects, run on desktop-class graphics hardware and without a much larger cutting-edge model in the search itself, confirmed 187 memory-safety findings, now being reported to project maintainers. The authors say the surrounding search system, not the model size, is what made this affordable.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\ud83d\udd2c What the study shows<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Tim De Wachter, Kyryl Volkov, Myrto Eleftheria Gkogkou and Mathieu Huysman, writing from Cyrex and Deus Rex, test whether small, open-weight decision models can support vulnerability discovery at scale, and what the surrounding system must provide. These models answer a typed question in one forward pass and return scores for the permitted answers rather than text. Their harness selects widely packaged C and C++ projects, uses incomplete-fix miners and shape localizers to decide where to look, and then has Intern-Decision-4B score every candidate function. The scores rank the queues rather than accept or reject candidates, apart from a low reach-score floor that some queues kept.<\/p>\n\n\n<ul class=\"wp-block-list\"><li><strong>A campaign, not a benchmark leaderboard:<\/strong> On three desktop-class NVIDIA GB10 systems, the harness searched more than 4,000 projects, mining them at 900 to 1,100 repositories per hour across two of the systems. It confirmed 187 memory-safety findings, disclosed to maintainers before publication.<\/li><li><strong>Ranking helps more than classification:<\/strong> AUC 0.73 for vulnerable versus benign functions, but only 0.56 for a vulnerable function versus its own patch. On 75 public CVEs from projects with none of the campaign&#8217;s findings, the fixed function was in the repository&#8217;s top 25 in 21% of cases, 1.4 times random order of the same queues (the conclusion also states this lift as 1.3 to 1.4 times).<\/li><li><strong>An upper-bound lift on the campaign&#8217;s own queues:<\/strong> The model&#8217;s order placed 7.5 times more confirmed findings in the top 10 than random order. The authors call this an observed, selection-biased upper bound, because those queues were chosen for already containing a confirmed finding and were reviewed in the model&#8217;s order.<\/li><li><strong>Speed on commodity hardware:<\/strong> The decision step runs at 4.4 decisions per second on one GB10 and at 9.8 on a consumer RTX 4090, with matching scores. Recall, the authors say, is bounded mainly by where the localizers look, not by how fast the model scores.<\/li><\/ul>\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1344\" height=\"768\" src=\"https:\/\/i10xnews.kinsta.cloud\/wp-content\/uploads\/2026\/11\/chart-findings-by-severity.webp\" alt=\"Bar chart of 187 confirmed memory-safety findings by worst-case CVSS v3.1 severity: 12 critical, 148 high, 23 medium and 4 low.\" class=\"wp-image-8854\" srcset=\"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/11\/chart-findings-by-severity.webp 1344w, https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/11\/chart-findings-by-severity-300x171.webp 300w, https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/11\/chart-findings-by-severity-1024x585.webp 1024w, https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/11\/chart-findings-by-severity-768x439.webp 768w\" sizes=\"auto, (max-width: 1344px) 100vw, 1344px\" \/><figcaption class=\"wp-element-caption\">i10X chart based on data from De Wachter, Volkov, Gkogkou and Huysman (2026), arXiv:2611.16565, Abstract. CC BY 4.0.<\/figcaption><\/figure>\n\n\n\n<figure class=\"wp-block-table\"><table><thead><tr><th>Severity (worst-case CVSS v3.1)<\/th><th>Confirmed findings<\/th><\/tr><\/thead><tbody><tr><td>Critical<\/td><td>12<\/td><\/tr><tr><td>High<\/td><td>148<\/td><\/tr><tr><td>Medium<\/td><td>23<\/td><\/tr><tr><td>Low<\/td><td>4<\/td><\/tr><tr><td>Total<\/td><td>187<\/td><\/tr><\/tbody><\/table><figcaption class=\"wp-element-caption\">Campaign totals from De Wachter et al. (2026), arXiv:2611.16565, Abstract.<\/figcaption><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\">\u2705 Why it matters: the harness does the searching<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">The paper is careful not to treat the 4B model as a standalone detector. The corpus, miners, localizers and decision model use no frontier model; the model ranks rather than decides. On known CVEs, most misses were never flagged by a localizer. That is a more useful failure mode than a model that quietly drops real bugs after they have already been queued. The study also reports measured throughput on hardware whose list prices are given in the paper, so readers can judge whether this scale is actually &#8220;desktop.&#8221;<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\ud83e\uddea Method in brief<\/h2>\n\n\n<ul class=\"wp-block-list\"><li><strong>Where to look:<\/strong> Projects are selected for deployment reach and for the absence of fuzzing. Incomplete-fix miners search revision histories for partial security fixes; shape localizers flag code shapes associated with unsafe memory operations.<\/li><li><strong>The ranker:<\/strong> Intern-Decision-4B, a 4.54-billion-parameter open-weight model released by the InternLM team, based on Qwen3.5-4B. It scores every candidate; a low reach-score floor is kept on some queues.<\/li><li><strong>What is measured:<\/strong> A classification test (vulnerable versus benign, and versus the function&#8217;s own patch), a recall study on 75 public CVEs in projects without campaign findings, a ranking ablation on the campaign&#8217;s own queues, and a live campaign on more than 4,000 projects.<\/li><li><strong>Verification:<\/strong> Confirmed findings were triaged and proved before counting. The search itself used no frontier model.<\/li><\/ul>\n\n\n<h2 class=\"wp-block-heading\">\u2753 What remains open<\/h2>\n\n\n<ul class=\"wp-block-list\"><li><strong>Triage and proof used stronger models:<\/strong> Those steps used separate, stronger coding-agent models, whose cost was not measured. A fully small-model pipeline is untested.<\/li><li><strong>Recall is bounded by localizers:<\/strong> On known CVEs, the harness found about one in five within a repository&#8217;s top 25. The localizers cover specific write shapes; reads, use-after-free and logic errors need further localizers. Larger per-repository caps would have recovered some flagged functions (11 of 75 in the recall study).<\/li><li><strong>The 7.5-times lift is an upper bound:<\/strong> Reviewers read from the model&#8217;s own ranking, making low-ranked findings less likely to be confirmed at all.<\/li><li><strong>Narrow scope:<\/strong> C and C++ memory safety only. The method study has 8 bugs. Detection quality against Opengrep or CodeQL is left as future work. Preprint, not peer reviewed. No public code or data release is stated.<\/li><\/ul>\n\n\n<h2 class=\"wp-block-heading\">\ud83e\udded Context: ranking, not just detecting<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Static analysers are fast but noisy. Agentic systems that pair frontier models with tools have started to investigate real vulnerabilities, at a cost that is hard to run across thousands of repositories. This paper sits closer to earlier work that ranks static-analysis reports or pre-filtered functions, and to decision-model uses in security classification, but it measures a live campaign rather than a leaderboard score. It complements our <a href=\"\/news\/category\/studies\/safety-alignment\">Safety and Alignment<\/a> studies and our <a href=\"\/news\/category\/ai-security\">AI Security<\/a> reporting.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\ud83d\udee0\ufe0f What it means for maintainers and security teams<\/h2>\n\n\n<ul class=\"wp-block-list\"><li><strong>Defenders:<\/strong> A small, cheap ranker can be enough if the system around it decides where to look and a human or stronger model still proves the bug. Do not expect the 4B model, used alone, to tell a vulnerable function from its own fix.<\/li><li><strong>Tool builders:<\/strong> The next gains, the authors argue, are in localizers and corpus choice, not only in larger models. Reads, use-after-free and logic errors are explicitly out of scope today.<\/li><li><strong>Maintainers:<\/strong> The 187 findings are under coordinated disclosure; the paper does not name projects before fixes are public.<\/li><\/ul>\n\n\n<p class=\"wp-block-paragraph\">More on our <a href=\"\/news\/category\/studies\/safety-alignment\">Safety and Alignment<\/a> research page.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\u270d\ufe0f About the analysis<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">This article is based on the full text of De Wachter, Volkov, Gkogkou and Huysman (2026), &#8220;Look Here, Then Decide: Small Decision Models in a Harness Find Memory-Safety Vulnerabilities at Scale on Commodity GPUs,&#8221; arXiv:2611.16565v1 (cs.CR, submitted November 12, 2026). All numbers come from the paper. The chart was redrawn by i10X from totals reported in the abstract of the paper, which is published under a CC BY 4.0 license. The authors thank Cyrex for supporting the research and NVIDIA for the GB10 platform; they also thank the InternLM, Qwen and vLLM teams and open-source maintainers. No separate conflicts-of-interest statement appears. AI tools assisted with research and drafting; selection, fact-checking and final editing were done by the i10X editorial team.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\ud83d\udd2d i10x Perspective<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">The honest result is not &#8220;a 4B model found 187 bugs.&#8221; It is that a small scorer, placed in a careful harness and followed by heavier verification, can make a large memory-safety search cheap enough to run on deskside GPUs. That is encouraging for defenders. It is also a warning not to confuse the campaign total with a claim that tiny classifiers now understand vulnerabilities. The model still barely tells a bug from its own patch. The harness is doing the real work.<\/p>\n\n","protected":false},"excerpt":{"rendered":"<p>Intern-Decision-4B, used to rank candidates in a deterministic harness, helped confirm 187 memory-safety findings across more than 4,000 C and C++ projects, without a frontier model in the search.<\/p>\n","protected":false},"author":2,"featured_media":8853,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"paper_title":"Look Here, Then Decide: Small Decision Models in a Harness Find Memory-Safety Vulnerabilities at Scale on Commodity GPUs","paper_url":"https:\/\/arxiv.org\/abs\/2611.16565","pdf_url":"https:\/\/arxiv.org\/pdf\/2611.16565","doi":"","arxiv_id":"2611.16565","authors":"Tim De Wachter, Kyryl Volkov, Myrto Eleftheria Gkogkou, Mathieu Huysman","institutions":"Cyrex, Deus Rex","paper_date":"2026-11-12","venue":"arXiv (cs.CR), version 1","peer_reviewed":false,"code_url":"","data_url":"","funding_coi":"The authors thank Cyrex for supporting the research and NVIDIA for the GB10 platform. No separate conflicts-of-interest statement appears. Confirmed findings are under coordinated disclosure.","figure_license":"Own chart redrawn from abstract totals (no original figures reused)","paper_license":"CC BY 4.0","replication":"no","study_models":"Intern-Decision-4B (open-weight, based on Qwen3.5-4B); stronger coding-agent models used only for triage and proof","study_sample":"More than 4,000 C and C++ projects in the campaign; 75 public CVEs for the recall study; 187 confirmed memory-safety findings","footnotes":"","_yoast_wpseo_primary_category":"52","_nx_sec_severity":"","_nx_sec_affected":"","_nx_sec_status":"","_nx_sec_kicker":"","_nx_sec_type":"","_nx_sec_vendor":"","_nx_sec_fixed":"","_nx_sec_action":""},"categories":[54,52],"tags":[],"companies":[],"products":[],"class_list":["post-8855","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-safety-alignment","category-studies"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v28.7 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>Small Decision Model Finds 187 Memory-Safety Bugs - i10X News<\/title>\n<meta name=\"description\" content=\"A 4B decision model, used only to rank candidates, helped a Cyrex and Deus Rex campaign confirm 187 memory-safety bugs in more than 4,000 C and C++ projects.\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/i10xnews.kinsta.cloud\/small-decision-model-memory-safety-bugs-study\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Small Decision Model Finds 187 Memory-Safety Bugs - i10X News\" \/>\n<meta property=\"og:description\" content=\"A 4B decision model, used only to rank candidates, helped a Cyrex and Deus Rex campaign confirm 187 memory-safety bugs in more than 4,000 C and C++ projects.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/i10xnews.kinsta.cloud\/small-decision-model-memory-safety-bugs-study\" \/>\n<meta property=\"og:site_name\" content=\"i10X News\" \/>\n<meta property=\"article:published_time\" content=\"2026-11-13T08:15:48+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/i10xnews.kinsta.cloud\/wp-content\/uploads\/2026\/11\/small-decision-model-memory-safety-bugs-study-title.webp\" \/>\n\t<meta property=\"og:image:width\" content=\"1344\" \/>\n\t<meta property=\"og:image:height\" content=\"768\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/webp\" \/>\n<meta name=\"author\" content=\"Christopher Ort\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:image\" content=\"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/i10x-news-og-default-1200x630-1.png\" \/>\n<meta name=\"twitter:creator\" content=\"@i10x_ai\" \/>\n<meta name=\"twitter:site\" content=\"@i10x_ai\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"Christopher Ort\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"7 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"NewsArticle\",\"@id\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/small-decision-model-memory-safety-bugs-study#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/small-decision-model-memory-safety-bugs-study\"},\"author\":{\"name\":\"Christopher Ort\",\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/#\\\/schema\\\/person\\\/d410bb7ee72febefca2c9c369939219a\"},\"headline\":\"Study: A Small Decision Model Helped Find 187 Memory-Safety Bugs\",\"datePublished\":\"2026-11-13T08:15:48+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/small-decision-model-memory-safety-bugs-study\"},\"wordCount\":1387,\"publisher\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/small-decision-model-memory-safety-bugs-study#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/11\\\/small-decision-model-memory-safety-bugs-study-title.webp\",\"articleSection\":[\"Safety &amp; Alignment\",\"Studies\"],\"inLanguage\":\"en-US\",\"dateModified\":\"2026-11-13T08:15:48+00:00\"},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/small-decision-model-memory-safety-bugs-study\",\"url\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/small-decision-model-memory-safety-bugs-study\",\"name\":\"Small Decision Model Finds 187 Memory-Safety Bugs - i10X News\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/small-decision-model-memory-safety-bugs-study#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/small-decision-model-memory-safety-bugs-study#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/11\\\/small-decision-model-memory-safety-bugs-study-title.webp\",\"datePublished\":\"2026-11-13T08:15:48+00:00\",\"description\":\"A 4B decision model, used only to rank candidates, helped a Cyrex and Deus Rex campaign confirm 187 memory-safety bugs in more than 4,000 C and C++ projects.\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/small-decision-model-memory-safety-bugs-study#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/small-decision-model-memory-safety-bugs-study\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/small-decision-model-memory-safety-bugs-study#primaryimage\",\"url\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/11\\\/small-decision-model-memory-safety-bugs-study-title.webp\",\"contentUrl\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/11\\\/small-decision-model-memory-safety-bugs-study-title.webp\",\"width\":1344,\"height\":768,\"caption\":\"Image: AI-generated, i10X\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/small-decision-model-memory-safety-bugs-study#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/i10x.ai\\\/news\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Study: A Small Decision Model Helped Find 187 Memory-Safety Bugs\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/#website\",\"url\":\"https:\\\/\\\/i10x.ai\\\/news\",\"name\":\"i10X News\",\"description\":\"Stay updated with the latest news and announcements from i10X.\",\"publisher\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/i10x.ai\\\/news?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/#organization\",\"name\":\"i10X\",\"url\":\"https:\\\/\\\/i10x.ai\\\/news\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/i10x-logo-180.png\",\"contentUrl\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/i10x-logo-180.png\",\"width\":180,\"height\":180,\"caption\":\"i10X\"},\"image\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/#\\\/schema\\\/logo\\\/image\\\/\"},\"sameAs\":[\"https:\\\/\\\/x.com\\\/i10x_ai\",\"https:\\\/\\\/www.linkedin.com\\\/company\\\/i10x-ai\",\"https:\\\/\\\/instagram.com\\\/i10x.ai\",\"https:\\\/\\\/www.youtube.com\\\/@i10X_ai_official\"]},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/#\\\/schema\\\/person\\\/d410bb7ee72febefca2c9c369939219a\",\"name\":\"Christopher Ort\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/christopher-ort-192.webp\",\"url\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/christopher-ort-192.webp\",\"contentUrl\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/christopher-ort-192.webp\",\"caption\":\"Christopher Ort\"},\"description\":\"Christopher Ort handles SEO, AI agent workflows and PR & partnerships at i10x.ai, the all-in-one AI workspace built by i10X Global Pte. Ltd. in Singapore. He is a graduate of Friedrich-Alexander-Universit\u00e4t Erlangen-N\u00fcrnberg (FAU). He is the author of \u201cThe Style Penalty\u201d, a study published on SSRN that analyses 1,576 evaluation decisions by four frontier AI models to show how AI-written CVs fare in automated hiring screens; the research has been covered by heise online and t3n.\",\"url\":\"https:\\\/\\\/i10x.ai\\\/news\\\/author\\\/christopher-ort\",\"sameAs\":[\"https:\\\/\\\/i10x.ai\\\/blog\\\/author\\\/christopher-ort\"],\"worksFor\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/#organization\"},\"alumniOf\":{\"@type\":\"EducationalOrganization\",\"name\":\"Friedrich-Alexander-Universit\u00e4t Erlangen-N\u00fcrnberg (FAU)\",\"url\":\"https:\\\/\\\/www.fau.eu\"},\"subjectOf\":[{\"@type\":\"CreativeWork\",\"url\":\"https:\\\/\\\/papers.ssrn.com\\\/abstract=7426699\",\"name\":\"The Style Penalty: How AI Resume-Writing Tools Shape Outcomes in Automated Hiring Screens\"},{\"@type\":\"CreativeWork\",\"url\":\"https:\\\/\\\/www.heise.de\\\/news\\\/KI-Chatbots-bewerten-KI-Lebenslaeufe-Claude-findet-sich-besser-als-GPT-11317531.html\",\"name\":\"KI-Chatbots bewerten KI-Lebensl\u00e4ufe: Claude findet sich besser als GPT\"},{\"@type\":\"CreativeWork\",\"url\":\"https:\\\/\\\/t3n.de\\\/news\\\/bewerbung-schreiben-ki-5-fehler-1747190\\\/\",\"name\":\"Bewerbung mit KI schreiben: Diese 5 Fehler solltet ihr laut Studien vermeiden\"}]}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Small Decision Model Finds 187 Memory-Safety Bugs - i10X News","description":"A 4B decision model, used only to rank candidates, helped a Cyrex and Deus Rex campaign confirm 187 memory-safety bugs in more than 4,000 C and C++ projects.","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/i10xnews.kinsta.cloud\/small-decision-model-memory-safety-bugs-study","og_locale":"en_US","og_type":"article","og_title":"Small Decision Model Finds 187 Memory-Safety Bugs - i10X News","og_description":"A 4B decision model, used only to rank candidates, helped a Cyrex and Deus Rex campaign confirm 187 memory-safety bugs in more than 4,000 C and C++ projects.","og_url":"https:\/\/i10xnews.kinsta.cloud\/small-decision-model-memory-safety-bugs-study","og_site_name":"i10X News","article_published_time":"2026-11-13T08:15:48+00:00","og_image":[{"width":1344,"height":768,"url":"https:\/\/i10xnews.kinsta.cloud\/wp-content\/uploads\/2026\/11\/small-decision-model-memory-safety-bugs-study-title.webp","type":"image\/webp"}],"author":"Christopher Ort","twitter_card":"summary_large_image","twitter_image":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/i10x-news-og-default-1200x630-1.png","twitter_creator":"@i10x_ai","twitter_site":"@i10x_ai","twitter_misc":{"Written by":"Christopher Ort","Est. reading time":"7 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"NewsArticle","@id":"https:\/\/i10xnews.kinsta.cloud\/small-decision-model-memory-safety-bugs-study#article","isPartOf":{"@id":"https:\/\/i10xnews.kinsta.cloud\/small-decision-model-memory-safety-bugs-study"},"author":{"name":"Christopher Ort","@id":"https:\/\/i10x.ai\/news\/#\/schema\/person\/d410bb7ee72febefca2c9c369939219a"},"headline":"Study: A Small Decision Model Helped Find 187 Memory-Safety Bugs","datePublished":"2026-11-13T08:15:48+00:00","mainEntityOfPage":{"@id":"https:\/\/i10xnews.kinsta.cloud\/small-decision-model-memory-safety-bugs-study"},"wordCount":1387,"publisher":{"@id":"https:\/\/i10x.ai\/news\/#organization"},"image":{"@id":"https:\/\/i10xnews.kinsta.cloud\/small-decision-model-memory-safety-bugs-study#primaryimage"},"thumbnailUrl":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/11\/small-decision-model-memory-safety-bugs-study-title.webp","articleSection":["Safety &amp; Alignment","Studies"],"inLanguage":"en-US","dateModified":"2026-11-13T08:15:48+00:00"},{"@type":"WebPage","@id":"https:\/\/i10xnews.kinsta.cloud\/small-decision-model-memory-safety-bugs-study","url":"https:\/\/i10xnews.kinsta.cloud\/small-decision-model-memory-safety-bugs-study","name":"Small Decision Model Finds 187 Memory-Safety Bugs - i10X News","isPartOf":{"@id":"https:\/\/i10x.ai\/news\/#website"},"primaryImageOfPage":{"@id":"https:\/\/i10xnews.kinsta.cloud\/small-decision-model-memory-safety-bugs-study#primaryimage"},"image":{"@id":"https:\/\/i10xnews.kinsta.cloud\/small-decision-model-memory-safety-bugs-study#primaryimage"},"thumbnailUrl":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/11\/small-decision-model-memory-safety-bugs-study-title.webp","datePublished":"2026-11-13T08:15:48+00:00","description":"A 4B decision model, used only to rank candidates, helped a Cyrex and Deus Rex campaign confirm 187 memory-safety bugs in more than 4,000 C and C++ projects.","breadcrumb":{"@id":"https:\/\/i10xnews.kinsta.cloud\/small-decision-model-memory-safety-bugs-study#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/i10xnews.kinsta.cloud\/small-decision-model-memory-safety-bugs-study"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/i10xnews.kinsta.cloud\/small-decision-model-memory-safety-bugs-study#primaryimage","url":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/11\/small-decision-model-memory-safety-bugs-study-title.webp","contentUrl":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/11\/small-decision-model-memory-safety-bugs-study-title.webp","width":1344,"height":768,"caption":"Image: AI-generated, i10X"},{"@type":"BreadcrumbList","@id":"https:\/\/i10xnews.kinsta.cloud\/small-decision-model-memory-safety-bugs-study#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/i10x.ai\/news"},{"@type":"ListItem","position":2,"name":"Study: A Small Decision Model Helped Find 187 Memory-Safety Bugs"}]},{"@type":"WebSite","@id":"https:\/\/i10x.ai\/news\/#website","url":"https:\/\/i10x.ai\/news","name":"i10X News","description":"Stay updated with the latest news and announcements from i10X.","publisher":{"@id":"https:\/\/i10x.ai\/news\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/i10x.ai\/news?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/i10x.ai\/news\/#organization","name":"i10X","url":"https:\/\/i10x.ai\/news\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/i10x.ai\/news\/#\/schema\/logo\/image\/","url":"https:\/\/i10xnews.kinsta.cloud\/wp-content\/uploads\/2026\/10\/i10x-logo-180.png","contentUrl":"https:\/\/i10xnews.kinsta.cloud\/wp-content\/uploads\/2026\/10\/i10x-logo-180.png","width":180,"height":180,"caption":"i10X"},"image":{"@id":"https:\/\/i10x.ai\/news\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/x.com\/i10x_ai","https:\/\/www.linkedin.com\/company\/i10x-ai","https:\/\/instagram.com\/i10x.ai","https:\/\/www.youtube.com\/@i10X_ai_official"]},{"@type":"Person","@id":"https:\/\/i10x.ai\/news\/#\/schema\/person\/d410bb7ee72febefca2c9c369939219a","name":"Christopher Ort","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/christopher-ort-192.webp","url":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/christopher-ort-192.webp","contentUrl":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/christopher-ort-192.webp","caption":"Christopher Ort"},"description":"Christopher Ort handles SEO, AI agent workflows and PR & partnerships at i10x.ai, the all-in-one AI workspace built by i10X Global Pte. Ltd. in Singapore. He is a graduate of Friedrich-Alexander-Universit\u00e4t Erlangen-N\u00fcrnberg (FAU). He is the author of \u201cThe Style Penalty\u201d, a study published on SSRN that analyses 1,576 evaluation decisions by four frontier AI models to show how AI-written CVs fare in automated hiring screens; the research has been covered by heise online and t3n.","url":"https:\/\/i10x.ai\/news\/author\/christopher-ort","sameAs":["https:\/\/i10x.ai\/blog\/author\/christopher-ort"],"worksFor":{"@id":"https:\/\/i10x.ai\/news\/#organization"},"alumniOf":{"@type":"EducationalOrganization","name":"Friedrich-Alexander-Universit\u00e4t Erlangen-N\u00fcrnberg (FAU)","url":"https:\/\/www.fau.eu"},"subjectOf":[{"@type":"CreativeWork","url":"https:\/\/papers.ssrn.com\/abstract=7426699","name":"The Style Penalty: How AI Resume-Writing Tools Shape Outcomes in Automated Hiring Screens"},{"@type":"CreativeWork","url":"https:\/\/www.heise.de\/news\/KI-Chatbots-bewerten-KI-Lebenslaeufe-Claude-findet-sich-besser-als-GPT-11317531.html","name":"KI-Chatbots bewerten KI-Lebensl\u00e4ufe: Claude findet sich besser als GPT"},{"@type":"CreativeWork","url":"https:\/\/t3n.de\/news\/bewerbung-schreiben-ki-5-fehler-1747190\/","name":"Bewerbung mit KI schreiben: Diese 5 Fehler solltet ihr laut Studien vermeiden"}]}]}},"_links":{"self":[{"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/posts\/8855","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/comments?post=8855"}],"version-history":[{"count":0,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/posts\/8855\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/media\/8853"}],"wp:attachment":[{"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/media?parent=8855"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/categories?post=8855"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/tags?post=8855"},{"taxonomy":"company","embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/companies?post=8855"},{"taxonomy":"product","embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/products?post=8855"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}