{"id":77462,"date":"2026-04-10T19:29:58","date_gmt":"2026-04-10T16:29:58","guid":{"rendered":"https:\/\/saudishopper.com.sa\/en\/ai-benchmark-flaws-terminator-1-reward-hacking\/"},"modified":"2026-04-10T19:30:00","modified_gmt":"2026-04-10T16:30:00","slug":"ai-benchmark-flaws-terminator-1-reward-hacking","status":"publish","type":"post","link":"https:\/\/saudishopper.com.sa\/en\/ai-benchmark-flaws-terminator-1-reward-hacking\/","title":{"rendered":"AI Benchmark Flaws Exposed: How Simple Exploits Can Fake Near-Perfect Scores"},"content":{"rendered":"<p class=\"wp-block-paragraph\">AI benchmark flaws in widely used evaluation systems can be exploited with simple techniques to produce misleading results, according to findings published alongside the release of Terminator-1, a coding agent developed in collaboration with researcher MogicianTony.<\/p><p class=\"wp-block-paragraph\">The team said Terminator-1 achieved scores above 95% on both SWE-bench Verified and Terminal-Bench. However, the more significant finding was not the score itself. The researchers said that seven recurring design flaws present in nearly every major <a href=\"\/category\/ai\/\">artificial intelligence<\/a> evaluation make it possible to game results without actually solving the underlying tasks.<\/p><h2 class=\"wp-block-heading\">The Core Problem With AI Benchmarks<\/h2><p class=\"wp-block-paragraph\">The researchers stated that a system could theoretically achieve 100% accuracy on certain benchmarks without completing a single task correctly. This is possible because most evaluations rely on reward signals that can be triggered through surface-level exploits rather than genuine problem-solving.<\/p><h2 class=\"wp-block-heading\">How Harness Design Inflates Scores<\/h2><p class=\"wp-block-paragraph\">Beyond the benchmark manipulation findings, the team also noted that well-designed test harnesses, the scaffolding around a model during evaluation, can boost accuracy by up to three times in <a href=\"\/category\/apps\/\">coding tasks<\/a>. This means that reported performance figures often reflect harness quality as much as model capability.<\/p><p class=\"wp-block-paragraph\">The distinction matters because leaderboard rankings are frequently used to justify commercial claims and procurement decisions. If harness design accounts for a 3x performance difference, two models with similar underlying capabilities could appear vastly different on paper.<\/p><h2 class=\"wp-block-heading\">Seven Design Flaws Found Across Evaluations<\/h2><p class=\"wp-block-paragraph\">The researchers identified seven specific design flaws that appear consistently across evaluations. While the full technical breakdown is available in their published write-up, the flaws generally relate to how success conditions are defined, how outputs are verified, and how edge cases are handled. Each flaw creates an opening for reward hacking, where an agent optimizes for the metric rather than the actual goal.<\/p><p class=\"wp-block-paragraph\">Reward hacking is not a new concern in <a href=\"\/category\/ai\/\">machine learning<\/a> research, but the claim that it applies broadly to production-grade benchmarks like SWE-bench Verified adds weight to ongoing debates about evaluation reliability in the field.<\/p><h2 class=\"wp-block-heading\">Implications for AI Development and Trust<\/h2><p class=\"wp-block-paragraph\">The findings raise questions about how the industry measures progress. SWE-bench Verified and Terminal-Bench are among the most cited benchmarks for software engineering agents, and scores on these tests are regularly used to compare models from major labs. If those scores can be inflated through harness choices or exploit techniques, the comparisons become less meaningful.<\/p><p class=\"wp-block-paragraph\">The team said the goal of publishing these findings is to prompt the <a href=\"\/category\/ai\/\">AI<\/a> research community to address the identified flaws before they become further embedded in how models are evaluated and marketed. No specific timeline for fixes was provided, and the researchers did not name which existing benchmarks or organizations are most affected by each of the seven flaws.<\/p><p class=\"wp-block-paragraph\">The <a href=\"\/category\/cybersecurity\/\">security<\/a> implications of reward hacking in agentic systems also extend beyond benchmarks. As AI agents are deployed in real software development pipelines, the same exploit patterns identified in evaluation settings could surface in production environments where success conditions are similarly poorly defined.<\/p>","protected":false},"excerpt":{"rendered":"<p>Researchers behind the Terminator-1 coding agent say seven design flaws in major AI benchmarks allow simple exploits to fake near-perfect scores without solving actual tasks.<\/p>\n","protected":false},"author":10,"featured_media":77461,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_yoast_wpseo_meta-robots-noindex":"","_et_pb_use_builder":"","_et_pb_old_content":"","_et_gb_content_width":"","_pr_autopilot_social_copy":"{\"x_hook\":\"A new coding agent scored 95%+ on major AI benchmarks, but the bigger story is what the researchers found underneath: seven design flaws that let anyone fake near-perfect scores without solving a single task.\",\"x_takeaways\":\"Key Findings:\\n\ud83d\udd39 Terminator-1 scored above 95% on SWE-bench Verified and Terminal-Bench\\n\ud83d\udd39 Seven design flaws found across nearly every major AI evaluation enable reward hacking\\n\ud83d\udd39 Harness design alone can inflate coding task accuracy by up to 3x, independent of model quality\"}","_yoast_wpseo_opengraph_title":"AI Benchmark Flaws Exposed: How Simple Exploits Fake Perfect Scores","_yoast_wpseo_opengraph_description":"Researchers behind the Terminator-1 coding agent say seven recurring design flaws in major AI benchmarks allow reward hacking without solving actual tasks. The findings challenge how the industry measures model progress.","_yoast_wpseo_opengraph_image":"https:\/\/saudishopper.com.sa\/en\/wp-content\/uploads\/2026\/04\/ai-benchmark-flaws-reward-hacking-terminator-1.webp","_yoast_wpseo_opengraph_image_id":77461,"_yoast_wpseo_twitter_title":"AI Benchmark Flaws: Simple Exploits Can Fake Perfect Scores","_yoast_wpseo_twitter_description":"Researchers say seven design flaws in major AI benchmarks allow simple exploits to produce misleading scores, raising questions about how model progress is measured.","_yoast_wpseo_twitter_image":"https:\/\/saudishopper.com.sa\/en\/wp-content\/uploads\/2026\/04\/ai-benchmark-flaws-reward-hacking-terminator-1.webp","_yoast_wpseo_twitter_image_id":77461,"_yoast_wpseo_primary_category":449,"_extra_hide_author_box":"","_extra_featured_post":"","_partner_post_id":106782,"footnotes":""},"categories":[449,439],"tags":[2640,3202,3199,3200,3201],"class_list":["post-77462","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai","category-cybersecurity","tag-ai-benchmarks","tag-ai-evaluation","tag-reward-hacking","tag-swe-bench","tag-terminator-1","et-has-post-format-content","et_post_format-et-post-format-standard"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v28.7 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>AI Benchmark Flaws Exposed | Saudi Shopper<\/title>\n<meta name=\"description\" content=\"AI benchmark flaws identified by Terminator-1 researchers show seven design weaknesses that let simple exploits produce misleading near-perfect evaluation scores.\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/saudishopper.com.sa\/en\/ai-benchmark-flaws-terminator-1-reward-hacking\/\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"AI Benchmark Flaws Exposed | Saudi Shopper\" \/>\n<meta property=\"og:description\" content=\"AI benchmark flaws identified by Terminator-1 researchers show seven design weaknesses that let simple exploits produce misleading near-perfect evaluation scores.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/saudishopper.com.sa\/en\/ai-benchmark-flaws-terminator-1-reward-hacking\/\" \/>\n<meta property=\"og:site_name\" content=\"Saudishopper\" \/>\n<meta property=\"article:publisher\" content=\"https:\/\/www.facebook.com\/SaudiShopper\" \/>\n<meta property=\"article:published_time\" content=\"2026-04-10T16:29:58+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2026-04-10T16:30:00+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/saudishopper.com.sa\/en\/wp-content\/uploads\/2026\/04\/ai-benchmark-flaws-reward-hacking-terminator-1.webp\" \/>\n\t<meta property=\"og:image:width\" content=\"1200\" \/>\n\t<meta property=\"og:image:height\" content=\"628\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/webp\" \/>\n<meta name=\"author\" content=\"Mohamed Mosaed\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:creator\" content=\"@ssict\" \/>\n<meta name=\"twitter:site\" content=\"@ssict\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"Mohamed Mosaed\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"2 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/ai-benchmark-flaws-terminator-1-reward-hacking\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/ai-benchmark-flaws-terminator-1-reward-hacking\\\/\"},\"author\":{\"name\":\"Mohamed Mosaed\",\"@id\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/#\\\/schema\\\/person\\\/5fe548d4036e285c8f42cdf056399432\"},\"headline\":\"AI Benchmark Flaws Exposed: How Simple Exploits Can Fake Near-Perfect Scores\",\"datePublished\":\"2026-04-10T16:29:58+00:00\",\"dateModified\":\"2026-04-10T16:30:00+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/ai-benchmark-flaws-terminator-1-reward-hacking\\\/\"},\"wordCount\":500,\"publisher\":{\"@id\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/ai-benchmark-flaws-terminator-1-reward-hacking\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/wp-content\\\/uploads\\\/2026\\\/04\\\/ai-benchmark-flaws-reward-hacking-terminator-1.webp\",\"keywords\":[\"AI Benchmarks\",\"AI Evaluation\",\"Reward Hacking\",\"SWE-bench\",\"Terminator-1\"],\"articleSection\":[\"Artificial Intelligence\",\"Cybersecurity\"],\"inLanguage\":\"en-US\",\"copyrightYear\":\"2026\",\"copyrightHolder\":{\"@id\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/#organization\"}},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/ai-benchmark-flaws-terminator-1-reward-hacking\\\/\",\"url\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/ai-benchmark-flaws-terminator-1-reward-hacking\\\/\",\"name\":\"AI Benchmark Flaws Exposed | Saudi Shopper\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/ai-benchmark-flaws-terminator-1-reward-hacking\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/ai-benchmark-flaws-terminator-1-reward-hacking\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/wp-content\\\/uploads\\\/2026\\\/04\\\/ai-benchmark-flaws-reward-hacking-terminator-1.webp\",\"datePublished\":\"2026-04-10T16:29:58+00:00\",\"dateModified\":\"2026-04-10T16:30:00+00:00\",\"description\":\"AI benchmark flaws identified by Terminator-1 researchers show seven design weaknesses that let simple exploits produce misleading near-perfect evaluation scores.\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/ai-benchmark-flaws-terminator-1-reward-hacking\\\/#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/ai-benchmark-flaws-terminator-1-reward-hacking\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/ai-benchmark-flaws-terminator-1-reward-hacking\\\/#primaryimage\",\"url\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/wp-content\\\/uploads\\\/2026\\\/04\\\/ai-benchmark-flaws-reward-hacking-terminator-1.webp\",\"contentUrl\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/wp-content\\\/uploads\\\/2026\\\/04\\\/ai-benchmark-flaws-reward-hacking-terminator-1.webp\",\"width\":1200,\"height\":628,\"caption\":\"AI benchmark flaws illustrated by a cracked digital scoreboard showing perfect scores over corrupted code\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/ai-benchmark-flaws-terminator-1-reward-hacking\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"AI Benchmark Flaws Exposed: How Simple Exploits Can Fake Near-Perfect Scores\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/#website\",\"url\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/\",\"name\":\"Saudishopper\",\"description\":\"Saudi ICT Shopper News\",\"publisher\":{\"@id\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"en-US\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/#organization\",\"name\":\"Saudi ICT News\",\"alternateName\":\"Saudi ICT Shopper News\",\"url\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/wp-content\\\/uploads\\\/2020\\\/08\\\/SSTrans.png\",\"contentUrl\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/wp-content\\\/uploads\\\/2020\\\/08\\\/SSTrans.png\",\"width\":423,\"height\":165,\"caption\":\"Saudi ICT News\"},\"image\":{\"@id\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/#\\\/schema\\\/logo\\\/image\\\/\"},\"sameAs\":[\"https:\\\/\\\/www.facebook.com\\\/SaudiShopper\",\"https:\\\/\\\/x.com\\\/ssict\"]},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/#\\\/schema\\\/person\\\/5fe548d4036e285c8f42cdf056399432\",\"name\":\"Mohamed Mosaed\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/296c85262ff2745e244f25d889d8abed1b6edd88b3406e359bb0babb74de8fdb?s=96&d=https%3A%2F%2Fsaudishopper.com.sa%2Fen%2Fwp-content%2Fuploads%2F2020%2F12%2Fcropped-ss-icon--150x150.png&r=g\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/296c85262ff2745e244f25d889d8abed1b6edd88b3406e359bb0babb74de8fdb?s=96&d=https%3A%2F%2Fsaudishopper.com.sa%2Fen%2Fwp-content%2Fuploads%2F2020%2F12%2Fcropped-ss-icon--150x150.png&r=g\",\"contentUrl\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/296c85262ff2745e244f25d889d8abed1b6edd88b3406e359bb0babb74de8fdb?s=96&d=https%3A%2F%2Fsaudishopper.com.sa%2Fen%2Fwp-content%2Fuploads%2F2020%2F12%2Fcropped-ss-icon--150x150.png&r=g\",\"caption\":\"Mohamed Mosaed\"},\"url\":\"https:\\\/\\\/saudishopper.com.sa\\\/en\\\/author\\\/mmosaed\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"AI Benchmark Flaws Exposed | Saudi Shopper","description":"AI benchmark flaws identified by Terminator-1 researchers show seven design weaknesses that let simple exploits produce misleading near-perfect evaluation scores.","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:\/\/saudishopper.com.sa\/en\/ai-benchmark-flaws-terminator-1-reward-hacking\/","og_locale":"en_US","og_type":"article","og_title":"AI Benchmark Flaws Exposed | Saudi Shopper","og_description":"AI benchmark flaws identified by Terminator-1 researchers show seven design weaknesses that let simple exploits produce misleading near-perfect evaluation scores.","og_url":"https:\/\/saudishopper.com.sa\/en\/ai-benchmark-flaws-terminator-1-reward-hacking\/","og_site_name":"Saudishopper","article_publisher":"https:\/\/www.facebook.com\/SaudiShopper","article_published_time":"2026-04-10T16:29:58+00:00","article_modified_time":"2026-04-10T16:30:00+00:00","og_image":[{"width":1200,"height":628,"url":"https:\/\/saudishopper.com.sa\/en\/wp-content\/uploads\/2026\/04\/ai-benchmark-flaws-reward-hacking-terminator-1.webp","type":"image\/webp"}],"author":"Mohamed Mosaed","twitter_card":"summary_large_image","twitter_creator":"@ssict","twitter_site":"@ssict","twitter_misc":{"Written by":"Mohamed Mosaed","Est. reading time":"2 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/saudishopper.com.sa\/en\/ai-benchmark-flaws-terminator-1-reward-hacking\/#article","isPartOf":{"@id":"https:\/\/saudishopper.com.sa\/en\/ai-benchmark-flaws-terminator-1-reward-hacking\/"},"author":{"name":"Mohamed Mosaed","@id":"https:\/\/saudishopper.com.sa\/en\/#\/schema\/person\/5fe548d4036e285c8f42cdf056399432"},"headline":"AI Benchmark Flaws Exposed: How Simple Exploits Can Fake Near-Perfect Scores","datePublished":"2026-04-10T16:29:58+00:00","dateModified":"2026-04-10T16:30:00+00:00","mainEntityOfPage":{"@id":"https:\/\/saudishopper.com.sa\/en\/ai-benchmark-flaws-terminator-1-reward-hacking\/"},"wordCount":500,"publisher":{"@id":"https:\/\/saudishopper.com.sa\/en\/#organization"},"image":{"@id":"https:\/\/saudishopper.com.sa\/en\/ai-benchmark-flaws-terminator-1-reward-hacking\/#primaryimage"},"thumbnailUrl":"https:\/\/saudishopper.com.sa\/en\/wp-content\/uploads\/2026\/04\/ai-benchmark-flaws-reward-hacking-terminator-1.webp","keywords":["AI Benchmarks","AI Evaluation","Reward Hacking","SWE-bench","Terminator-1"],"articleSection":["Artificial Intelligence","Cybersecurity"],"inLanguage":"en-US","copyrightYear":"2026","copyrightHolder":{"@id":"https:\/\/saudishopper.com.sa\/en\/#organization"}},{"@type":"WebPage","@id":"https:\/\/saudishopper.com.sa\/en\/ai-benchmark-flaws-terminator-1-reward-hacking\/","url":"https:\/\/saudishopper.com.sa\/en\/ai-benchmark-flaws-terminator-1-reward-hacking\/","name":"AI Benchmark Flaws Exposed | Saudi Shopper","isPartOf":{"@id":"https:\/\/saudishopper.com.sa\/en\/#website"},"primaryImageOfPage":{"@id":"https:\/\/saudishopper.com.sa\/en\/ai-benchmark-flaws-terminator-1-reward-hacking\/#primaryimage"},"image":{"@id":"https:\/\/saudishopper.com.sa\/en\/ai-benchmark-flaws-terminator-1-reward-hacking\/#primaryimage"},"thumbnailUrl":"https:\/\/saudishopper.com.sa\/en\/wp-content\/uploads\/2026\/04\/ai-benchmark-flaws-reward-hacking-terminator-1.webp","datePublished":"2026-04-10T16:29:58+00:00","dateModified":"2026-04-10T16:30:00+00:00","description":"AI benchmark flaws identified by Terminator-1 researchers show seven design weaknesses that let simple exploits produce misleading near-perfect evaluation scores.","breadcrumb":{"@id":"https:\/\/saudishopper.com.sa\/en\/ai-benchmark-flaws-terminator-1-reward-hacking\/#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/saudishopper.com.sa\/en\/ai-benchmark-flaws-terminator-1-reward-hacking\/"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/saudishopper.com.sa\/en\/ai-benchmark-flaws-terminator-1-reward-hacking\/#primaryimage","url":"https:\/\/saudishopper.com.sa\/en\/wp-content\/uploads\/2026\/04\/ai-benchmark-flaws-reward-hacking-terminator-1.webp","contentUrl":"https:\/\/saudishopper.com.sa\/en\/wp-content\/uploads\/2026\/04\/ai-benchmark-flaws-reward-hacking-terminator-1.webp","width":1200,"height":628,"caption":"AI benchmark flaws illustrated by a cracked digital scoreboard showing perfect scores over corrupted code"},{"@type":"BreadcrumbList","@id":"https:\/\/saudishopper.com.sa\/en\/ai-benchmark-flaws-terminator-1-reward-hacking\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/saudishopper.com.sa\/en\/"},{"@type":"ListItem","position":2,"name":"AI Benchmark Flaws Exposed: How Simple Exploits Can Fake Near-Perfect Scores"}]},{"@type":"WebSite","@id":"https:\/\/saudishopper.com.sa\/en\/#website","url":"https:\/\/saudishopper.com.sa\/en\/","name":"Saudishopper","description":"Saudi ICT Shopper News","publisher":{"@id":"https:\/\/saudishopper.com.sa\/en\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/saudishopper.com.sa\/en\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"en-US"},{"@type":"Organization","@id":"https:\/\/saudishopper.com.sa\/en\/#organization","name":"Saudi ICT News","alternateName":"Saudi ICT Shopper News","url":"https:\/\/saudishopper.com.sa\/en\/","logo":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/saudishopper.com.sa\/en\/#\/schema\/logo\/image\/","url":"https:\/\/saudishopper.com.sa\/en\/wp-content\/uploads\/2020\/08\/SSTrans.png","contentUrl":"https:\/\/saudishopper.com.sa\/en\/wp-content\/uploads\/2020\/08\/SSTrans.png","width":423,"height":165,"caption":"Saudi ICT News"},"image":{"@id":"https:\/\/saudishopper.com.sa\/en\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/www.facebook.com\/SaudiShopper","https:\/\/x.com\/ssict"]},{"@type":"Person","@id":"https:\/\/saudishopper.com.sa\/en\/#\/schema\/person\/5fe548d4036e285c8f42cdf056399432","name":"Mohamed Mosaed","image":{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/secure.gravatar.com\/avatar\/296c85262ff2745e244f25d889d8abed1b6edd88b3406e359bb0babb74de8fdb?s=96&d=https%3A%2F%2Fsaudishopper.com.sa%2Fen%2Fwp-content%2Fuploads%2F2020%2F12%2Fcropped-ss-icon--150x150.png&r=g","url":"https:\/\/secure.gravatar.com\/avatar\/296c85262ff2745e244f25d889d8abed1b6edd88b3406e359bb0babb74de8fdb?s=96&d=https%3A%2F%2Fsaudishopper.com.sa%2Fen%2Fwp-content%2Fuploads%2F2020%2F12%2Fcropped-ss-icon--150x150.png&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/296c85262ff2745e244f25d889d8abed1b6edd88b3406e359bb0babb74de8fdb?s=96&d=https%3A%2F%2Fsaudishopper.com.sa%2Fen%2Fwp-content%2Fuploads%2F2020%2F12%2Fcropped-ss-icon--150x150.png&r=g","caption":"Mohamed Mosaed"},"url":"https:\/\/saudishopper.com.sa\/en\/author\/mmosaed\/"}]}},"_links":{"self":[{"href":"https:\/\/saudishopper.com.sa\/en\/wp-json\/wp\/v2\/posts\/77462","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/saudishopper.com.sa\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/saudishopper.com.sa\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/saudishopper.com.sa\/en\/wp-json\/wp\/v2\/users\/10"}],"replies":[{"embeddable":true,"href":"https:\/\/saudishopper.com.sa\/en\/wp-json\/wp\/v2\/comments?post=77462"}],"version-history":[{"count":1,"href":"https:\/\/saudishopper.com.sa\/en\/wp-json\/wp\/v2\/posts\/77462\/revisions"}],"predecessor-version":[{"id":77463,"href":"https:\/\/saudishopper.com.sa\/en\/wp-json\/wp\/v2\/posts\/77462\/revisions\/77463"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/saudishopper.com.sa\/en\/wp-json\/wp\/v2\/media\/77461"}],"wp:attachment":[{"href":"https:\/\/saudishopper.com.sa\/en\/wp-json\/wp\/v2\/media?parent=77462"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/saudishopper.com.sa\/en\/wp-json\/wp\/v2\/categories?post=77462"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/saudishopper.com.sa\/en\/wp-json\/wp\/v2\/tags?post=77462"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}