{"id":6100,"date":"2026-10-19T16:48:48","date_gmt":"2026-10-19T16:48:48","guid":{"rendered":"https:\/\/i10xnews.kinsta.cloud\/?p=6100"},"modified":"2026-10-19T17:14:42","modified_gmt":"2026-10-19T17:14:42","slug":"behavioral-shadows-post-training-taskless-distillation","status":"publish","type":"post","link":"https:\/\/i10x.ai\/news\/behavioral-shadows-post-training-taskless-distillation","title":{"rendered":"A Model&#8217;s Coding Skill Can Leak Through Single Unrelated Words, Study Finds"},"content":{"rendered":"\n<h2 class=\"wp-block-heading\">\u26a1 Quick Take<\/h2>\n\n<ul class=\"wp-block-list\"><li><strong>The finding:<\/strong> A student model that never saw coding examples gained 5.34 percentage points on HumanEval+ by learning only from 5,664 single-word answers the teacher gave on unrelated prompts.<\/li><li><strong>Not a fluke:<\/strong> An exact nuisance-matched control that shuffled prompt\u2013response pairings wiped out the gain. The effect also appeared in science, commonsense and reading tasks.<\/li><li><strong>What it means:<\/strong> Post-training leaves a behavioral shadow on ordinary word choices. That shadow can transmit capabilities \u2014 and, in principle, be observed without access to the updated weights.<\/li><\/ul>\n\n\n<p class=\"wp-block-paragraph\">Post-training is usually judged by the task it targets: a coding update should improve code. Researchers from Peking University, Georgia Tech, ShanghaiTech, Tsinghua and Lovart AI ask whether the update also changes which ordinary word a model prefers in a short story that contains no code. They call those changes the behavioral shadow of post-training.<\/p>\n\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\/10\/behavioral-shadows-post-training-taskless-distillation-humaneval.webp\" alt=\"Column chart: nuisance-matched control 46.19 percent, ATD student 51.22 percent matching the teacher, a 5.34-point gain from single-word training alone.\" class=\"wp-image-6099\" srcset=\"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/behavioral-shadows-post-training-taskless-distillation-humaneval.webp 1344w, https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/behavioral-shadows-post-training-taskless-distillation-humaneval-300x171.webp 300w, https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/behavioral-shadows-post-training-taskless-distillation-humaneval-1024x585.webp 1024w, https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/behavioral-shadows-post-training-taskless-distillation-humaneval-768x439.webp 768w\" sizes=\"auto, (max-width: 1344px) 100vw, 1344px\" \/><figcaption class=\"wp-element-caption\">i10X chart based on data from Zhang et al. (2026), arXiv:2609.29233, Table 1 (primary coding experiment) (CC BY 4.0).<\/figcaption><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\">\ud83d\udd2c What the study shows<\/h2>\n\n\n<p class=\"wp-block-paragraph\">Active Taskless Distillation (ATD) finds prompts where a shared public ancestor model is nearly indifferent between two ordinary words. Near that tie, a private post-training update can flip the teacher&#8217;s choice. The student, initialized from the same ancestor, learns only from those prompt\u2013word pairs \u2014 no target-task examples, teacher logits or teacher parameters.<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table><thead><tr><th>Setting (Qwen2.5-1.5B coding)<\/th><th>HumanEval+<\/th><\/tr><\/thead><tbody><tr><td>Teacher (post-trained)<\/td><td>51.22%<\/td><\/tr><tr><td>ATD student (single-word only)<\/td><td>51.22%<\/td><\/tr><tr><td>Exact nuisance-matched control<\/td><td>45.88\u201346.19%<\/td><\/tr><tr><td>Student \u2212 control<\/td><td>+5.34 pp [1.22, 9.60]<\/td><\/tr><\/tbody><\/table><figcaption class=\"wp-element-caption\">Primary coding experiment; 95% bootstrap CI as reported.<\/figcaption><\/figure>\n\n\n<ul class=\"wp-block-list\"><li><strong>Matches the teacher:<\/strong> The student reached the teacher&#8217;s aggregate coding score without coding data.<\/li><li><strong>Broad transfer:<\/strong> Gains of 0.81\u20135.03 pp appeared across ScienceQA, OpenBookQA, HellaSwag, RACE, CommonsenseQA and PIQA.<\/li><li><strong>Source-specific and composable:<\/strong> The learned signal tracked the teacher&#8217;s update strength and could be composed across sources.<\/li><li><strong>Active queries matter:<\/strong> Near-boundary prompts beat passive sampling at the same query budget by 4.27 pp on HumanEval+.<\/li><li><strong>Cross-family:<\/strong> Positive mean gains appeared across additional model generations, sizes and families, including Llama-3.2-1B and Qwen3 variants.<\/li><\/ul>\n\n\n<h2 class=\"wp-block-heading\">\u2705 Why it means two things at once<\/h2>\n\n\n<p class=\"wp-block-paragraph\">For capability transfer, ATD is a minimal distillation recipe: one bit per prompt can move the needle. For security and IP, it is a warning that private post-training can be partially observed and copied through unrelated generations.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">It extends subliminal-learning results from traits and preferences to measurable task capabilities, using far less teacher text.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\ud83e\uddea Method in brief<\/h2>\n\n\n<p class=\"wp-block-paragraph\">Primary experiment: Qwen2.5-1.5B lineage, 5,664 single-word teacher responses. Controls include exact nuisance-matched permutation of pairings, global label shuffle and teacher-free public labels. Intervals are 95% bootstraps of paired signal-minus-control differences. Code: github.com\/myboker\/ATD.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\u2753 What remains open<\/h2>\n\n<ul class=\"wp-block-list\"><li><strong>Small models:<\/strong> Main results are on 1\u20134B-class models.<\/li><li><strong>Constructed near-ties:<\/strong> Queries are chosen where the ancestor is nearly indifferent; arbitrary prompts may leak less.<\/li><li><strong>Dual use:<\/strong> The same method can aid distillation or probing of private updates.<\/li><\/ul>\n\n\n<h2 class=\"wp-block-heading\">\ud83e\udded Context<\/h2>\n\n\n<p class=\"wp-block-paragraph\">Builds on subliminal learning work showing traits can pass through unrelated teacher outputs. Related safety and evaluation research is in <a href=\"\/news\/category\/studies\/safety-alignment\">Safety &amp; Alignment<\/a>.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\ud83d\udee0\ufe0f What it means for labs and auditors<\/h2>\n\n<ul class=\"wp-block-list\"><li><strong>Assume leakage:<\/strong> Do not treat unrelated generations as information-free about a private update.<\/li><li><strong>Audit with near-ties:<\/strong> Indifference-region probes are a practical observability tool.<\/li><li><strong>Control distillation carefully:<\/strong> Single-word corpora can still move benchmark scores.<\/li><\/ul>\n\n\n<h2 class=\"wp-block-heading\">\u270d\ufe0f About the analysis<\/h2>\n\n\n<p class=\"wp-block-paragraph\">This article is based on the full text of the preprint &#8220;Post-Training Leaves Behavioral Shadows on Unrelated Decisions&#8221; (arXiv:2609.29233v1, submitted September 24, 2026). All numbers are taken from the paper&#8217;s text, tables and figures; i10X did not rerun the experiments. The chart was redrawn by i10X from Table 1 (primary coding experiment) of the paper, which is published under a CC BY 4.0 license.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\ud83d\udd2d i10x Perspective<\/h2>\n\n\n<p class=\"wp-block-paragraph\">The striking detail is the minimalism: one ordinary word per prompt, no coding text, and the student still climbs HumanEval+. That collapses the comforting story that only task data matters. Behavioral shadows are now a first-class object for both distillers and anyone trying to keep post-training private.<\/p>\n\n","protected":false},"excerpt":{"rendered":"<p>Active Taskless Distillation: 5,664 single-word teacher replies transfer a 5.34-point HumanEval+ gain to a student that never saw coding examples.<\/p>\n","protected":false},"author":2,"featured_media":6250,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"paper_title":"Post-Training Leaves Behavioral Shadows on Unrelated Decisions","paper_url":"https:\/\/arxiv.org\/abs\/2609.29233","pdf_url":"https:\/\/arxiv.org\/pdf\/2609.29233v1","doi":"","arxiv_id":"2609.29233","authors":"Ziyang Zhang, Yubin Jing, Yuanhao Zeng, Yuyao Li, Haofan Wang, Yichen Gong","institutions":"Peking University; Georgia Institute of Technology; ShanghaiTech University; Tsinghua University; Lovart AI","paper_date":"2026-09-24","venue":"arXiv (cs.CL), version 1","peer_reviewed":false,"code_url":"https:\/\/github.com\/myboker\/ATD","data_url":"","funding_coi":"Affiliations include Lovart AI.","figure_license":"Chart drawn by i10X from Table 1 (primary coding experiment) (CC BY 4.0)","paper_license":"CC BY 4.0","replication":"no","study_models":"Qwen2.5-1.5B primary; also Llama-3.2-1B, Qwen3-1.7B\/4B and other lineages","study_sample":"5,664 single-word teacher responses (coding); transfer across multiple NLU benchmarks","footnotes":"","_yoast_wpseo_primary_category":""},"categories":[54,52],"tags":[],"companies":[],"products":[],"class_list":["post-6100","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>Behavioral Shadows: Post-Training Leaks Through Unrelated Words - i10X News<\/title>\n<meta name=\"description\" content=\"Study: coding ability transfers via single-word answers on unrelated prompts. 5.34-point HumanEval+ gain without any coding examples.\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/i10x.ai\/news\/behavioral-shadows-post-training-taskless-distillation\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Behavioral Shadows: Post-Training Leaks Through Unrelated Words - i10X News\" \/>\n<meta property=\"og:description\" content=\"Study: coding ability transfers via single-word answers on unrelated prompts. 5.34-point HumanEval+ gain without any coding examples.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/i10x.ai\/news\/behavioral-shadows-post-training-taskless-distillation\" \/>\n<meta property=\"og:site_name\" content=\"i10X News\" \/>\n<meta property=\"article:published_time\" content=\"2026-10-19T16:48:48+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2026-10-19T17:14:42+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/behavioral-shadows-post-training-taskless-distillation-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: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=\"3 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"NewsArticle\",\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/behavioral-shadows-post-training-taskless-distillation#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/behavioral-shadows-post-training-taskless-distillation\"},\"author\":{\"name\":\"Christopher Ort\",\"@id\":\"https:\\\/\\\/news.i10x.ai\\\/#\\\/schema\\\/person\\\/d410bb7ee72febefca2c9c369939219a\"},\"headline\":\"A Model&#8217;s Coding Skill Can Leak Through Single Unrelated Words, Study Finds\",\"datePublished\":\"2026-10-19T16:48:48+00:00\",\"dateModified\":\"2026-10-19T17:14:42+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/behavioral-shadows-post-training-taskless-distillation\"},\"wordCount\":654,\"commentCount\":0,\"publisher\":{\"@id\":\"https:\\\/\\\/news.i10x.ai\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/behavioral-shadows-post-training-taskless-distillation#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/behavioral-shadows-post-training-taskless-distillation-title.webp\",\"articleSection\":[\"Safety &amp; Alignment\",\"Studies\"],\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/i10x.ai\\\/news\\\/behavioral-shadows-post-training-taskless-distillation#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/behavioral-shadows-post-training-taskless-distillation\",\"url\":\"https:\\\/\\\/i10x.ai\\\/news\\\/behavioral-shadows-post-training-taskless-distillation\",\"name\":\"Behavioral Shadows: Post-Training Leaks Through Unrelated Words - i10X News\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/news.i10x.ai\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/behavioral-shadows-post-training-taskless-distillation#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/behavioral-shadows-post-training-taskless-distillation#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/behavioral-shadows-post-training-taskless-distillation-title.webp\",\"datePublished\":\"2026-10-19T16:48:48+00:00\",\"dateModified\":\"2026-10-19T17:14:42+00:00\",\"description\":\"Study: coding ability transfers via single-word answers on unrelated prompts. 5.34-point HumanEval+ gain without any coding examples.\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/behavioral-shadows-post-training-taskless-distillation#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/i10x.ai\\\/news\\\/behavioral-shadows-post-training-taskless-distillation\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/behavioral-shadows-post-training-taskless-distillation#primaryimage\",\"url\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/behavioral-shadows-post-training-taskless-distillation-title.webp\",\"contentUrl\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/behavioral-shadows-post-training-taskless-distillation-title.webp\",\"width\":1344,\"height\":768,\"caption\":\"Illustration: i10X\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/behavioral-shadows-post-training-taskless-distillation#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/news.i10x.ai\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"A Model&#8217;s Coding Skill Can Leak Through Single Unrelated Words, Study Finds\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/news.i10x.ai\\\/#website\",\"url\":\"https:\\\/\\\/news.i10x.ai\\\/\",\"name\":\"i10X News\",\"description\":\"Stay updated with the latest news and announcements from i10X.\",\"publisher\":{\"@id\":\"https:\\\/\\\/news.i10x.ai\\\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/news.i10x.ai\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/news.i10x.ai\\\/#organization\",\"name\":\"i10X\",\"url\":\"https:\\\/\\\/news.i10x.ai\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/news.i10x.ai\\\/#\\\/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:\\\/\\\/news.i10x.ai\\\/#\\\/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:\\\/\\\/news.i10x.ai\\\/#\\\/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\"}}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"Behavioral Shadows: Post-Training Leaks Through Unrelated Words - i10X News","description":"Study: coding ability transfers via single-word answers on unrelated prompts. 5.34-point HumanEval+ gain without any coding examples.","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:\/\/i10x.ai\/news\/behavioral-shadows-post-training-taskless-distillation","og_locale":"en_US","og_type":"article","og_title":"Behavioral Shadows: Post-Training Leaks Through Unrelated Words - i10X News","og_description":"Study: coding ability transfers via single-word answers on unrelated prompts. 5.34-point HumanEval+ gain without any coding examples.","og_url":"https:\/\/i10x.ai\/news\/behavioral-shadows-post-training-taskless-distillation","og_site_name":"i10X News","article_published_time":"2026-10-19T16:48:48+00:00","article_modified_time":"2026-10-19T17:14:42+00:00","og_image":[{"width":1344,"height":768,"url":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/behavioral-shadows-post-training-taskless-distillation-title.webp","type":"image\/webp"}],"author":"Christopher Ort","twitter_card":"summary_large_image","twitter_creator":"@i10x_ai","twitter_site":"@i10x_ai","twitter_misc":{"Written by":"Christopher Ort","Est. reading time":"3 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"NewsArticle","@id":"https:\/\/i10x.ai\/news\/behavioral-shadows-post-training-taskless-distillation#article","isPartOf":{"@id":"https:\/\/i10x.ai\/news\/behavioral-shadows-post-training-taskless-distillation"},"author":{"name":"Christopher Ort","@id":"https:\/\/news.i10x.ai\/#\/schema\/person\/d410bb7ee72febefca2c9c369939219a"},"headline":"A Model&#8217;s Coding Skill Can Leak Through Single Unrelated Words, Study Finds","datePublished":"2026-10-19T16:48:48+00:00","dateModified":"2026-10-19T17:14:42+00:00","mainEntityOfPage":{"@id":"https:\/\/i10x.ai\/news\/behavioral-shadows-post-training-taskless-distillation"},"wordCount":654,"commentCount":0,"publisher":{"@id":"https:\/\/news.i10x.ai\/#organization"},"image":{"@id":"https:\/\/i10x.ai\/news\/behavioral-shadows-post-training-taskless-distillation#primaryimage"},"thumbnailUrl":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/behavioral-shadows-post-training-taskless-distillation-title.webp","articleSection":["Safety &amp; Alignment","Studies"],"inLanguage":"en-US","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/i10x.ai\/news\/behavioral-shadows-post-training-taskless-distillation#respond"]}]},{"@type":"WebPage","@id":"https:\/\/i10x.ai\/news\/behavioral-shadows-post-training-taskless-distillation","url":"https:\/\/i10x.ai\/news\/behavioral-shadows-post-training-taskless-distillation","name":"Behavioral Shadows: Post-Training Leaks Through Unrelated Words - i10X News","isPartOf":{"@id":"https:\/\/news.i10x.ai\/#website"},"primaryImageOfPage":{"@id":"https:\/\/i10x.ai\/news\/behavioral-shadows-post-training-taskless-distillation#primaryimage"},"image":{"@id":"https:\/\/i10x.ai\/news\/behavioral-shadows-post-training-taskless-distillation#primaryimage"},"thumbnailUrl":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/behavioral-shadows-post-training-taskless-distillation-title.webp","datePublished":"2026-10-19T16:48:48+00:00","dateModified":"2026-10-19T17:14:42+00:00","description":"Study: coding ability transfers via single-word answers on unrelated prompts. 5.34-point HumanEval+ gain without any coding examples.","breadcrumb":{"@id":"https:\/\/i10x.ai\/news\/behavioral-shadows-post-training-taskless-distillation#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/i10x.ai\/news\/behavioral-shadows-post-training-taskless-distillation"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/i10x.ai\/news\/behavioral-shadows-post-training-taskless-distillation#primaryimage","url":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/behavioral-shadows-post-training-taskless-distillation-title.webp","contentUrl":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/behavioral-shadows-post-training-taskless-distillation-title.webp","width":1344,"height":768,"caption":"Illustration: i10X"},{"@type":"BreadcrumbList","@id":"https:\/\/i10x.ai\/news\/behavioral-shadows-post-training-taskless-distillation#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/news.i10x.ai\/"},{"@type":"ListItem","position":2,"name":"A Model&#8217;s Coding Skill Can Leak Through Single Unrelated Words, Study Finds"}]},{"@type":"WebSite","@id":"https:\/\/news.i10x.ai\/#website","url":"https:\/\/news.i10x.ai\/","name":"i10X News","description":"Stay updated with the latest news and announcements from i10X.","publisher":{"@id":"https:\/\/news.i10x.ai\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/news.i10x.ai\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/news.i10x.ai\/#organization","name":"i10X","url":"https:\/\/news.i10x.ai\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/news.i10x.ai\/#\/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:\/\/news.i10x.ai\/#\/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:\/\/news.i10x.ai\/#\/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"}}]}},"_links":{"self":[{"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/posts\/6100","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=6100"}],"version-history":[{"count":2,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/posts\/6100\/revisions"}],"predecessor-version":[{"id":6251,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/posts\/6100\/revisions\/6251"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/media\/6250"}],"wp:attachment":[{"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/media?parent=6100"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/categories?post=6100"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/tags?post=6100"},{"taxonomy":"company","embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/companies?post=6100"},{"taxonomy":"product","embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/products?post=6100"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}