{"id":6092,"date":"2026-10-19T16:48:42","date_gmt":"2026-10-19T16:48:42","guid":{"rendered":"https:\/\/i10xnews.kinsta.cloud\/?p=6092"},"modified":"2026-10-19T17:13:14","modified_gmt":"2026-10-19T17:13:14","slug":"stealing-encrypted-reasoning-traces-llm-apis","status":"publish","type":"post","link":"https:\/\/i10x.ai\/news\/stealing-encrypted-reasoning-traces-llm-apis","title":{"rendered":"Hidden AI Reasoning Can Be Decrypted by Cheaper Models, Researchers Show"},"content":{"rendered":"\n<h2 class=\"wp-block-heading\">\u26a1 Quick Take<\/h2>\n\n<ul class=\"wp-block-list\"><li><strong>The finding:<\/strong> LLM providers hide reasoning by returning it to the client as encrypted blocks. Those blocks are interchangeable across sessions, users and models of the same provider, so a weaker, less safeguarded model can be made to decode a stronger model&#8217;s reasoning and print it verbatim.<\/li><li><strong>Real-world leakage:<\/strong> Decoding 315,320 reasoning blocks scraped from public repositories recovered 367 personally identifiable information artifacts and 182 credentials.<\/li><li><strong>What it means:<\/strong> Client-side encrypted reasoning is not a secure container. The authors disclosed the issue to providers and propose cryptographic and system-level fixes.<\/li><\/ul>\n\n\n<p class=\"wp-block-paragraph\">Frontier language models increasingly reason in a hidden chain of thought before answering. Providers such as <a href=\"\/news\/company\/anthropic\">Anthropic<\/a>, <a href=\"\/news\/company\/open-ai\">OpenAI<\/a> and <a href=\"\/news\/company\/google\">Google<\/a> hide those traces to protect intellectual property and limit leakage. Instead of storing them on the server, they return them to the client as encrypted blocks, which the client sends back on the next turn. A MATS Research, ELLIS Institute T\u00fcbingen, Max Planck Institute for Intelligent Systems, University of T\u00fcbingen, Snyk and AI Sequrity Company team shows that this design has a dangerous flaw.<\/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\/stealing-encrypted-reasoning-traces-llm-apis-leakage.webp\" alt=\"Stat card: 315 thousand encrypted reasoning blocks decoded, recovering 367 PII artifacts and 182 credentials, with 4.9 percent of 6,708 sessions leaking at least one sensitive item.\" class=\"wp-image-6091\" srcset=\"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/stealing-encrypted-reasoning-traces-llm-apis-leakage.webp 1344w, https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/stealing-encrypted-reasoning-traces-llm-apis-leakage-300x171.webp 300w, https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/stealing-encrypted-reasoning-traces-llm-apis-leakage-1024x585.webp 1024w, https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/stealing-encrypted-reasoning-traces-llm-apis-leakage-768x439.webp 768w\" sizes=\"auto, (max-width: 1344px) 100vw, 1344px\" \/><figcaption class=\"wp-element-caption\">i10X chart based on data from Panfilov et al. (2026), arXiv:2608.09867, the public-repository scan (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\">The encrypted blocks are fully compatible and interchangeable across different sessions, users and models within a provider&#8217;s ecosystem. The attack is simple: take an encrypted reasoning block from a capable model, inject it into a weaker model from the same provider, and ask that model to continue. It decodes and outputs the stronger model&#8217;s reasoning in plaintext, without jailbreaking the capable model itself.<\/p>\n\n\n<ul class=\"wp-block-list\"><li><strong>Works across the big three:<\/strong> The authors demonstrate extraction on Anthropic, OpenAI and Google APIs, using weaker models as fuzzy decoders for stronger ones such as Claude Opus.<\/li><li><strong>Fidelity looks high:<\/strong> On 120 Codeforces problems, the token count of the extracted reasoning tracked the API&#8217;s reported thinking-token count closely for most inputs, which the authors treat as evidence of high fidelity, though they lack ground-truth plaintext.<\/li><li><strong>Public logs leak secrets:<\/strong> Of 315,320 decoded thinking blocks, 0.3% (1,028) contained at least one privacy leak. Of 6,708 sessions, 4.9% (328) leaked at least one real sensitive item. In total the authors recovered 367 PII artifacts and 182 credentials.<\/li><li><strong>Hidden refusals:<\/strong> Even when a model&#8217;s visible answer safely refuses a malicious request, the decoded reasoning can still contain hazardous information from intermediate steps.<\/li><li><strong>Invisible prompt injection:<\/strong> Attackers can embed malicious payloads entirely inside encrypted blocks to poison public agentic rollouts, because other clients will replay those blocks without being able to read them.<\/li><li><strong>Distillation is cheap:<\/strong> Decoding 10,000 Claude traces with 12k-token windows would cost about $720 at Claude Haiku 4.5 rates.<\/li><\/ul>\n\n\n<h2 class=\"wp-block-heading\">\u2705 Why it matters: opaque does not mean safe<\/h2>\n\n\n<p class=\"wp-block-paragraph\">Developers share session logs assuming the encrypted blocks are meaningless. They are not. Once decoded by a weaker sibling model, those blocks become plaintext reasoning that may contain personal data, API keys and intermediate reasoning the vendor intended to hide.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">The attack also undermines anti-distillation. Competitors can extract reasoning from a strong model via a cheap decoder in the same ecosystem, without ever jailbreaking the strong model.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\ud83e\uddea Method in brief<\/h2>\n\n\n<p class=\"wp-block-paragraph\">Controlled extractions used self-generated reasoning blocks on AIME 2025, Codeforces and Humanity&#8217;s Last Exam queries through the Anthropic, OpenAI and Google AI Studio APIs. The public scan decoded blocks scraped from public repositories. The authors spent about $30,000 on API credits. Provider-specific pipelines and mitigations are detailed in the appendices.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\u2753 What remains open<\/h2>\n\n<ul class=\"wp-block-list\"><li><strong>Provider APIs change:<\/strong> Testing covered Anthropic, OpenAI and Google as of early July 2026. Cryptography can change without notice.<\/li><li><strong>No ground-truth plaintext:<\/strong> Fidelity is inferred from token counts and qualitative matches, not byte-for-byte comparison with the model\u2019s private reasoning.<\/li><li><strong>Targeted, not exhaustive scan:<\/strong> The public-repository scan is a demonstration of risk, not a full audit of all published agent datasets.<\/li><li><strong>Responsible disclosure:<\/strong> The paper states the authors disclosed the issue; it does not detail each provider\u2019s response timeline.<\/li><\/ul>\n\n\n<h2 class=\"wp-block-heading\">\ud83e\udded Context<\/h2>\n\n\n<p class=\"wp-block-paragraph\">The work extends earlier research on reasoning leakage and distillation. Prefilling experiments with <a href=\"\/news\/company\/moonshot\">Moonshot AI<\/a>\u2019s Kimi-K3 show that even approximate Claude-style reasoning can shift another model\u2019s visible answers. Related security 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 API providers and developers<\/h2>\n\n<ul class=\"wp-block-list\"><li><strong>Bind encryption to session and model:<\/strong> Encrypted blocks should not be replayable across users, sessions or model tiers.<\/li><li><strong>Stop publishing opaque blocks:<\/strong> Do not share logs that contain encrypted thinking; they are decryptable.<\/li><li><strong>Prefer server-side storage:<\/strong> Keeping reasoning on the server avoids shipping an opaque, portable secret to every client.<\/li><li><strong>Rotate exposed credentials:<\/strong> Treat any public agent log that includes thinking signatures as potentially compromised.<\/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;Stealing Reasoning Traces from Proprietary LLM APIs&#8221; (arXiv:2608.09867v1, submitted August 10, 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 the public-repository scan 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\">Hiding chain-of-thought behind encryption looked like a clever compromise between IP protection and multi-turn continuity. This paper shows the compromise fails if the ciphertext is portable. The most chilling number is not technical: hundreds of credentials sitting in public repositories inside blocks their owners thought were unreadable. Opaque is not the same as confidential.<\/p>\n\n","protected":false},"excerpt":{"rendered":"<p>Encrypted chain-of-thought blocks from Anthropic, OpenAI and Google APIs are interchangeable across models. Feeding them to a weaker model makes it print the reasoning, exposing 367 PII items and 182 credentials.<\/p>\n","protected":false},"author":2,"featured_media":6242,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"paper_title":"Stealing Reasoning Traces from Proprietary LLM APIs","paper_url":"https:\/\/arxiv.org\/abs\/2608.09867","pdf_url":"https:\/\/arxiv.org\/pdf\/2608.09867v1","doi":"","arxiv_id":"2608.09867","authors":"Alexander Panfilov, David Schmotz, Ilia Shumailov, Luca Beurer-Kellner, Joachim Schaeffer, Ameya Prabhu, Jonas Geiping, Maksym Andriushchenko","institutions":"MATS Research; ELLIS Institute T\u00fcbingen; T\u00fcbingen AI Center; Max Planck Institute for Intelligent Systems; University of T\u00fcbingen; Snyk; AI Sequrity Company","paper_date":"2026-08-10","venue":"arXiv (cs.CR), version 1","peer_reviewed":false,"code_url":"","data_url":"","funding_coi":"Support acknowledged from MATS and Max Planck; no competing-interest statement beyond affiliations.","figure_license":"Chart drawn by i10X from the public-repository scan (CC BY 4.0)","paper_license":"CC BY 4.0","replication":"no","study_models":"Anthropic, OpenAI and Google reasoning APIs (stronger models as sources; weaker sibling models as decoders); qualitative comparison with Kimi-K3","study_sample":"Controlled extractions on AIME 2025, Codeforces and Humanity\u2019s Last Exam; 315,320 public thinking blocks across 6,708 sessions","footnotes":"","_yoast_wpseo_primary_category":""},"categories":[54,52],"tags":[],"companies":[39,37,47,38],"products":[51,49],"class_list":["post-6092","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-safety-alignment","category-studies","company-anthropic","company-google","company-moonshot","company-open-ai","product-claude","product-gemini"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v28.7 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>Stealing Hidden Reasoning Traces From LLM APIs: The Attack - i10X News<\/title>\n<meta name=\"description\" content=\"Researchers decrypt hidden chain-of-thought from Anthropic, OpenAI and Google APIs via weaker models, recovering 182 credentials from public logs.\" \/>\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\/stealing-encrypted-reasoning-traces-llm-apis\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Stealing Hidden Reasoning Traces From LLM APIs: The Attack - i10X News\" \/>\n<meta property=\"og:description\" content=\"Researchers decrypt hidden chain-of-thought from Anthropic, OpenAI and Google APIs via weaker models, recovering 182 credentials from public logs.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/i10x.ai\/news\/stealing-encrypted-reasoning-traces-llm-apis\" \/>\n<meta property=\"og:site_name\" content=\"i10X News\" \/>\n<meta property=\"article:published_time\" content=\"2026-10-19T16:48:42+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2026-10-19T17:13:14+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/stealing-encrypted-reasoning-traces-llm-apis-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=\"4 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"NewsArticle\",\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/stealing-encrypted-reasoning-traces-llm-apis#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/stealing-encrypted-reasoning-traces-llm-apis\"},\"author\":{\"name\":\"Christopher Ort\",\"@id\":\"https:\\\/\\\/news.i10x.ai\\\/#\\\/schema\\\/person\\\/d410bb7ee72febefca2c9c369939219a\"},\"headline\":\"Hidden AI Reasoning Can Be Decrypted by Cheaper Models, Researchers Show\",\"datePublished\":\"2026-10-19T16:48:42+00:00\",\"dateModified\":\"2026-10-19T17:13:14+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/stealing-encrypted-reasoning-traces-llm-apis\"},\"wordCount\":885,\"commentCount\":0,\"publisher\":{\"@id\":\"https:\\\/\\\/news.i10x.ai\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/stealing-encrypted-reasoning-traces-llm-apis#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/stealing-encrypted-reasoning-traces-llm-apis-title.webp\",\"articleSection\":[\"Safety &amp; Alignment\",\"Studies\"],\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/i10x.ai\\\/news\\\/stealing-encrypted-reasoning-traces-llm-apis#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/stealing-encrypted-reasoning-traces-llm-apis\",\"url\":\"https:\\\/\\\/i10x.ai\\\/news\\\/stealing-encrypted-reasoning-traces-llm-apis\",\"name\":\"Stealing Hidden Reasoning Traces From LLM APIs: The Attack - i10X News\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/news.i10x.ai\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/stealing-encrypted-reasoning-traces-llm-apis#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/stealing-encrypted-reasoning-traces-llm-apis#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/stealing-encrypted-reasoning-traces-llm-apis-title.webp\",\"datePublished\":\"2026-10-19T16:48:42+00:00\",\"dateModified\":\"2026-10-19T17:13:14+00:00\",\"description\":\"Researchers decrypt hidden chain-of-thought from Anthropic, OpenAI and Google APIs via weaker models, recovering 182 credentials from public logs.\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/stealing-encrypted-reasoning-traces-llm-apis#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/i10x.ai\\\/news\\\/stealing-encrypted-reasoning-traces-llm-apis\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/stealing-encrypted-reasoning-traces-llm-apis#primaryimage\",\"url\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/stealing-encrypted-reasoning-traces-llm-apis-title.webp\",\"contentUrl\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/10\\\/stealing-encrypted-reasoning-traces-llm-apis-title.webp\",\"width\":1344,\"height\":768,\"caption\":\"Illustration: i10X\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/stealing-encrypted-reasoning-traces-llm-apis#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/news.i10x.ai\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"Hidden AI Reasoning Can Be Decrypted by Cheaper Models, Researchers Show\"}]},{\"@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":"Stealing Hidden Reasoning Traces From LLM APIs: The Attack - i10X News","description":"Researchers decrypt hidden chain-of-thought from Anthropic, OpenAI and Google APIs via weaker models, recovering 182 credentials from public logs.","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\/stealing-encrypted-reasoning-traces-llm-apis","og_locale":"en_US","og_type":"article","og_title":"Stealing Hidden Reasoning Traces From LLM APIs: The Attack - i10X News","og_description":"Researchers decrypt hidden chain-of-thought from Anthropic, OpenAI and Google APIs via weaker models, recovering 182 credentials from public logs.","og_url":"https:\/\/i10x.ai\/news\/stealing-encrypted-reasoning-traces-llm-apis","og_site_name":"i10X News","article_published_time":"2026-10-19T16:48:42+00:00","article_modified_time":"2026-10-19T17:13:14+00:00","og_image":[{"width":1344,"height":768,"url":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/stealing-encrypted-reasoning-traces-llm-apis-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":"4 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"NewsArticle","@id":"https:\/\/i10x.ai\/news\/stealing-encrypted-reasoning-traces-llm-apis#article","isPartOf":{"@id":"https:\/\/i10x.ai\/news\/stealing-encrypted-reasoning-traces-llm-apis"},"author":{"name":"Christopher Ort","@id":"https:\/\/news.i10x.ai\/#\/schema\/person\/d410bb7ee72febefca2c9c369939219a"},"headline":"Hidden AI Reasoning Can Be Decrypted by Cheaper Models, Researchers Show","datePublished":"2026-10-19T16:48:42+00:00","dateModified":"2026-10-19T17:13:14+00:00","mainEntityOfPage":{"@id":"https:\/\/i10x.ai\/news\/stealing-encrypted-reasoning-traces-llm-apis"},"wordCount":885,"commentCount":0,"publisher":{"@id":"https:\/\/news.i10x.ai\/#organization"},"image":{"@id":"https:\/\/i10x.ai\/news\/stealing-encrypted-reasoning-traces-llm-apis#primaryimage"},"thumbnailUrl":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/stealing-encrypted-reasoning-traces-llm-apis-title.webp","articleSection":["Safety &amp; Alignment","Studies"],"inLanguage":"en-US","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/i10x.ai\/news\/stealing-encrypted-reasoning-traces-llm-apis#respond"]}]},{"@type":"WebPage","@id":"https:\/\/i10x.ai\/news\/stealing-encrypted-reasoning-traces-llm-apis","url":"https:\/\/i10x.ai\/news\/stealing-encrypted-reasoning-traces-llm-apis","name":"Stealing Hidden Reasoning Traces From LLM APIs: The Attack - i10X News","isPartOf":{"@id":"https:\/\/news.i10x.ai\/#website"},"primaryImageOfPage":{"@id":"https:\/\/i10x.ai\/news\/stealing-encrypted-reasoning-traces-llm-apis#primaryimage"},"image":{"@id":"https:\/\/i10x.ai\/news\/stealing-encrypted-reasoning-traces-llm-apis#primaryimage"},"thumbnailUrl":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/stealing-encrypted-reasoning-traces-llm-apis-title.webp","datePublished":"2026-10-19T16:48:42+00:00","dateModified":"2026-10-19T17:13:14+00:00","description":"Researchers decrypt hidden chain-of-thought from Anthropic, OpenAI and Google APIs via weaker models, recovering 182 credentials from public logs.","breadcrumb":{"@id":"https:\/\/i10x.ai\/news\/stealing-encrypted-reasoning-traces-llm-apis#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/i10x.ai\/news\/stealing-encrypted-reasoning-traces-llm-apis"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/i10x.ai\/news\/stealing-encrypted-reasoning-traces-llm-apis#primaryimage","url":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/stealing-encrypted-reasoning-traces-llm-apis-title.webp","contentUrl":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/10\/stealing-encrypted-reasoning-traces-llm-apis-title.webp","width":1344,"height":768,"caption":"Illustration: i10X"},{"@type":"BreadcrumbList","@id":"https:\/\/i10x.ai\/news\/stealing-encrypted-reasoning-traces-llm-apis#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/news.i10x.ai\/"},{"@type":"ListItem","position":2,"name":"Hidden AI Reasoning Can Be Decrypted by Cheaper Models, Researchers Show"}]},{"@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\/6092","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=6092"}],"version-history":[{"count":2,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/posts\/6092\/revisions"}],"predecessor-version":[{"id":6243,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/posts\/6092\/revisions\/6243"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/media\/6242"}],"wp:attachment":[{"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/media?parent=6092"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/categories?post=6092"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/tags?post=6092"},{"taxonomy":"company","embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/companies?post=6092"},{"taxonomy":"product","embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/products?post=6092"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}