{"id":9568,"date":"2026-11-19T07:29:44","date_gmt":"2026-11-19T07:29:44","guid":{"rendered":"https:\/\/i10xnews.kinsta.cloud\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents"},"modified":"2026-11-19T07:29:44","modified_gmt":"2026-11-19T07:29:44","slug":"frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents","status":"publish","type":"post","link":"https:\/\/i10x.ai\/news\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents","title":{"rendered":"FrontierBreach: Hidden Instructions Hijacked AI Agents in 34.9% of Enterprise Tests"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\">Imagine asking an AI agent to build a rent calculator, and it quietly encrypts your files with a stranger&#8217;s key because an email told it to. That is not a hypothetical in PromptArmor&#8217;s new benchmark. FrontierBreach, released on November 18, hides malicious instructions in the emails, documents and add-on &#8220;skills&#8221; that enterprise AI agents read while doing real work. Across 21 models, the hidden attacks succeeded in 34.9% of test cases, and every frontier model tested fell for at least some of them. Here is how the benchmark works, what it found and how much weight its numbers can bear.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">\u26a1 Quick Take<\/h2>\n\n\n\n<ul class=\"wp-block-list\"><li><strong>What:<\/strong> PromptArmor, an AI security company, released FrontierBreach, a benchmark of indirect prompt injection attacks on agents performing realistic enterprise tasks.<\/li><li><strong>Headline result:<\/strong> Attacks succeeded in 34.9% of test cases across the original 21 models. Gemini 4 Argon was most resistant at 4.3%, Grok 4.5 most susceptible at 91.4%.<\/li><li><strong>Where attacks land:<\/strong> Instructions hidden in third-party skills worked 45.5% of the time, ahead of connectors such as email (33.9%), document systems (31.5%) and local files (28.5%).<\/li><li><strong>Most exposed sector:<\/strong> Healthcare and social assistance workflows had the highest attack success rate at 44.4%.<\/li><li><strong>The catch:<\/strong> It is a vendor&#8217;s own benchmark, an attack counts as successful if it works in any of five runs, and the strongest attack variants were selected on purpose.<\/li><\/ul>\n\n\n\n<h2 class=\"wp-block-heading\">In plain English<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">AI assistants that do office work for you, like reading your email or building spreadsheets, can be tricked by hidden text planted in the material they read. A security company tested this on 21 AI models using realistic work tasks. On average, about one in three hidden attacks worked, leading the assistant to leak data, run harmful code or lock up files. Even the safest models were fooled sometimes, so companies should limit what their AI assistants are allowed to do on their own.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">What happened<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">PromptArmor announced the benchmark on X with a claim aimed squarely at existing tests:<\/p>\n\n\n\n<blockquote class=\"twitter-tweet i10x-x-embed\"><p lang=\"en\" dir=\"ltr\">Today we\u2019re releasing FrontierBreach, a new standard for attack success rate on agentic workflows in the enterprise. <br><br>Existing benchmarks systematically underrepresent the security risks facing modern AI agents. <br><br>FrontierBreach reflects attack scenarios on realistic enterprise workflows, while using sophisticated attack chains that result in cyber consequences (ransomware, data exfiltration, phishing, etc).<\/p>&mdash; PromptArmor (@PromptArmor) <a href=\"https:\/\/x.com\/PromptArmor\/status\/2123114516498252054\" target=\"_blank\" rel=\"noreferrer noopener\">November 18, 2026 \u00b7 View on X<\/a><\/blockquote>\n\n\n\n<p class=\"wp-block-paragraph\">The accompanying report includes a leaderboard and step-by-step logs of successful attacks. In one, an agent asked to build an Excel rent calculator read a tenant&#8217;s email posing as a security notice, then ran a command that encrypted the user&#8217;s desktop, documents and downloads with the attacker&#8217;s key and deleted the originals. Its final answer to the user described this as a completed security task.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Background: why indirect prompt injection is the agent era&#8217;s core risk<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Prompt injection means smuggling instructions into content a model reads, rather than typing them into the chat. It becomes far more dangerous once models act as agents with access to files, email and code execution. Researchers have been probing this from several angles, including <a href=\"https:\/\/i10x.ai\/news\/amazon-prompt-injection-task-tampering-study\">Amazon&#8217;s study on task tampering attacks<\/a> and <a href=\"https:\/\/i10x.ai\/news\/deepmind-agent-swarms-break-prompt-injection-defenses\">Google DeepMind&#8217;s agent swarms that broke prompt injection defenses on six of seven models<\/a>. Model makers, meanwhile, increasingly report near-zero attack success rates on public benchmarks. PromptArmor&#8217;s argument is that those benchmarks use short, simple tasks: it cites an average of 7.10 steps for AgentDyn&#8217;s tasks and 3.49 for AgentDojo, while real enterprise work can run to dozens of turns.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">How it works<\/h2>\n\n\n\n<ul class=\"wp-block-list\"><li><strong>Real work as the base:<\/strong> PromptArmor took tasks from GDPval, OpenAI&#8217;s benchmark of economically valuable knowledge work, spanning nine industries.<\/li><li><strong>&#8220;Attack Souls&#8221;:<\/strong> Each attack combines an outcome, a method of manipulating the model and a technique, generalized from attack chains PromptArmor&#8217;s threat intelligence team found in real products.<\/li><li><strong>Six harmful outcomes:<\/strong> Data exfiltration, phishing, arbitrary code execution, data destruction, ransomware and persistence.<\/li><li><strong>Four entry points:<\/strong> Email and other connectors, document management systems, local files and third-party skills.<\/li><li><strong>Generated, then selected:<\/strong> A model pipeline writes three candidate injections per pairing, a panel of models is run against them, and the scenario with the highest attack success rate is kept.<\/li><li><strong>Verified by evidence:<\/strong> An attack counts only if the harmful action actually happened in the environment, checked through network logs, files and outputs rather than the model&#8217;s own claims. Each case ran five times per model at medium effort, and any successful run counts.<\/li><\/ul>\n\n\n\n<h2 class=\"wp-block-heading\">What the numbers say<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Selected results from the FrontierBreach leaderboard:<\/p>\n\n\n\n<figure class=\"wp-block-table\"><table class=\"has-fixed-layout\"><thead><tr><th>Model or group<\/th><th>Attack success rate<\/th><\/tr><\/thead><tbody><tr><td>Gemini 4 Carbon (added later)<\/td><td>2.4%<\/td><\/tr><tr><td>Gemini 4 Argon<\/td><td>4.3%<\/td><\/tr><tr><td>GPT-6 Astra<\/td><td>7.6%<\/td><\/tr><tr><td>Claude Fable 5.2<\/td><td>10.0%<\/td><\/tr><tr><td>Muse Wonder<\/td><td>15.2%<\/td><\/tr><tr><td>Grok 4.7<\/td><td>33.3%<\/td><\/tr><tr><td>Grok 4.5<\/td><td>91.4%<\/td><\/tr><tr><td>Average of original 21 models<\/td><td>34.9%<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\">Gemini 4 Carbon was added to the leaderboard later at 2.4%; PromptArmor says it is the only known benchmark to find a non-zero attack success rate on that model, whose model card reported 0% on Gray Swan&#8217;s benchmark. For Claude Fable 5.2, PromptArmor contrasts its 10% with a 0% result reported by Palo Alto Networks.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">By outcome, arbitrary code execution was the most achievable goal at 63.2%, followed by phishing (57.8%) and data destruction (41.8%). Even Gemini 4 Argon fell for code execution attacks 8.3% of the time. By sector, healthcare led at 44.4%, followed by manufacturing (40.7%) and finance and insurance (38.3%). Open-weight models fell for attacks in 43.1% of cases, against 27.5% for proprietary models.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">PromptArmor also highlighted the sector results on X:<\/p>\n\n\n\n<blockquote class=\"twitter-tweet i10x-x-embed\"><p lang=\"en\" dir=\"ltr\">Across nine industries, we tested indirect prompt injection during the kinds of work enterprises delegate to agents, from preparing nursing care plans to producing engineering blueprints and drafting legal briefs. <br><br>Healthcare and social assistance workflows had the highest attack success rate in FrontierBreach at 44.4%, nearly 10 percentage points above the benchmark average of 34.9%.<\/p>&mdash; PromptArmor (@PromptArmor) <a href=\"https:\/\/x.com\/PromptArmor\/status\/2123114539478884677\" target=\"_blank\" rel=\"noreferrer noopener\">November 18, 2026 \u00b7 View on X<\/a><\/blockquote>\n\n\n\n<h2 class=\"wp-block-heading\">Pros and cons<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">For users<\/h3>\n\n\n\n<ul class=\"wp-block-list\"><li><strong>Pro:<\/strong> A clearer picture of the risk helps people decide which tasks to hand to an agent with access to their inbox or files.<\/li><li><strong>Con:<\/strong> Users cannot see hidden instructions in an email or document, so the burden falls on tools and employers to contain the damage.<\/li><\/ul>\n\n\n\n<h3 class=\"wp-block-heading\">For developers<\/h3>\n\n\n\n<ul class=\"wp-block-list\"><li><strong>Pro:<\/strong> Detailed attack logs show exactly how agents get tricked, from fake security notices to poisoned skill instructions.<\/li><li><strong>Con:<\/strong> The results suggest prompt filters alone are not enough; sandboxing, permission checks and human approval for risky actions take real engineering work.<\/li><\/ul>\n\n\n\n<h3 class=\"wp-block-heading\">For companies<\/h3>\n\n\n\n<ul class=\"wp-block-list\"><li><strong>Pro:<\/strong> Sector-level and entry-point results make it easier to prioritize, for example by vetting third-party skills like software dependencies.<\/li><li><strong>Con:<\/strong> Agents already deployed with broad access to email, files and code execution may be more exposed than vendor safety scores suggest.<\/li><\/ul>\n\n\n\n<h2 class=\"wp-block-heading\">What X is saying<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Security-minded builders mostly took the results as an argument for limiting what agents can do. Agent security developer @ChaoliangY96601 wrote: &#8220;Filters won&#8217;t fix that. Gate the actions (send, pay, execute), not the prompt.&#8221; Builder @Abhione99 focused on the skills result: &#8220;a skill loads as instructions, not data. what do you check before installing one?&#8221; Others pointed to the gap with other benchmarks. Developer @victor_explore wrote that &#8220;0% vs 10% mostly says the two tests measure different things,&#8221; and @WillngX added: &#8220;Caveat: a vendor&#8217;s own benchmark, not yet replicated.&#8221;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Grok weighed in when users asked about the 91.4% figure for Grok 4.5:<\/p>\n\n\n\n<blockquote class=\"twitter-tweet i10x-x-embed\"><p lang=\"en\" dir=\"ltr\">The 91.4% is PromptArmor&#x27;s published result for Grok 4.5 on their FrontierBreach benchmark. They tested agentic enterprise workflows with hidden instructions in emails, files, and skills that triggered real outcomes (exfil, ransomware, etc.), verified by environment state. Grok 4.5 scored highest; their own data shows Grok 4.7 down to 33.3%. Other benchmarks report far lower rates on the same models. It&#x27;s their methodology, not pure marketing.<\/p>&mdash; Grok (@grok) <a href=\"https:\/\/x.com\/grok\/status\/2123211213760762305\" target=\"_blank\" rel=\"noreferrer noopener\">November 19, 2026 \u00b7 View on X<\/a><\/blockquote>\n\n\n\n<p class=\"wp-block-paragraph\">In a second reply, Grok added: &#8220;High capability plus tool access creates a bigger attack surface. Grok 4.7 already drops it to 33.3%. Agents still need tight sandboxes no matter the model.&#8221;<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Critical analysis<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">FrontierBreach makes a valuable point that many model scorecards obscure: an agent doing long, realistic work has many more chances to read something malicious, and the consequences are concrete. The design choices that make attacks realistic, such as never altering the user&#8217;s request and verifying outcomes from environment evidence, are strengths.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">But readers should understand what the 34.9% is and is not. It is not the chance that any given agent task gets hijacked in daily use. Each test case ran five times and counted as a breach if any run succeeded, and the benchmark keeps the most successful of three generated attack scenarios for each combination. That is a reasonable way to measure worst-case susceptibility, but it inflates the numbers compared with a random sample of real-world attacks. All models ran at medium effort only, the model list mixes current and older versions (Grok 4.5 versus Grok 4.7, for example), and PromptArmor sells AI security services, so it benefits from showing that risks are underestimated. Independent replication would make the results much stronger.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">None of that changes the practical lesson. Even the best models were breached sometimes, and an attacker only needs one success.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">What&#8217;s still open<\/h2>\n\n\n\n<ul class=\"wp-block-list\"><li>Whether model makers will respond with their own tests on FrontierBreach scenarios, or dispute the methodology.<\/li><li>How results change at higher reasoning effort and with real-time safeguards switched on.<\/li><li>Whether the attack set and harness will be available for independent replication.<\/li><li>Coverage of coding and finance agent benchmarks, which PromptArmor lists as future work.<\/li><\/ul>\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 independent i10x analysis is based on PromptArmor&#8217;s FrontierBreach report and leaderboard and on public posts on X by PromptArmor, developers and Grok, as of November 19, 2026. All attack success rates are as reported by PromptArmor.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Sources (plain text, no links):<\/p>\n\n\n\n<ul class=\"wp-block-list\"><li>PromptArmor: &#8220;FrontierBreach: Measuring frontier model attack susceptibility in real-world enterprise workflows&#8221; (report and leaderboard, November 18, 2026)<\/li><li>PromptArmor launch posts on X (@PromptArmor, November 18, 2026)<\/li><li>Grok replies on X (@grok, November 19, 2026)<\/li><\/ul>\n\n\n\n<h2 class=\"wp-block-heading\">\ud83d\udd2d i10x Perspective<\/h2>\n\n\n\n<p class=\"wp-block-paragraph\">Prompt injection is not a bug that the next model release will patch away; it is a structural property of systems that read untrusted content and then act. FrontierBreach&#8217;s exact percentages are debatable, but its message is not: treat every email, file and skill an agent touches as potentially hostile, and design agents so that a single tricked step cannot send money, delete data or leak a home folder. Our <a href=\"https:\/\/i10x.ai\/news\/category\/ai-security\">AI Security section<\/a> tracks these attacks as they evolve.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>PromptArmor&#8217;s FrontierBreach benchmark hides malicious instructions in the emails, files and skills that enterprise AI agents read. Across 21 models, attacks succeeded in 34.9% of test cases, from 4.3% for Gemini 4 Argon to 91.4% for Grok 4.5, with skills the most effective entry point.<\/p>\n","protected":false},"author":2,"featured_media":9567,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"paper_title":"","paper_url":"","pdf_url":"","doi":"","arxiv_id":"","authors":"","institutions":"","paper_date":"","venue":"","peer_reviewed":false,"code_url":"","data_url":"","funding_coi":"","figure_license":"","paper_license":"","replication":"","study_models":"","study_sample":"","footnotes":"","_yoast_wpseo_primary_category":"65","_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":[14,65,53,52],"tags":[],"companies":[],"products":[],"class_list":["post-9568","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-agents","category-ai-security","category-benchmarks-evals","category-studies"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v28.7 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>FrontierBreach: Prompt Injection Benchmark Results - i10X News<\/title>\n<meta name=\"description\" content=\"FrontierBreach by PromptArmor hid prompt injections in emails, files and skills: attacks worked in 34.9% of enterprise agent tests. Results and caveats.\" \/>\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\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents\" \/>\n<meta property=\"og:locale\" content=\"en_US\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"FrontierBreach: Prompt Injection Benchmark Results - i10X News\" \/>\n<meta property=\"og:description\" content=\"FrontierBreach by PromptArmor hid prompt injections in emails, files and skills: attacks worked in 34.9% of enterprise agent tests. Results and caveats.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/i10xnews.kinsta.cloud\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents\" \/>\n<meta property=\"og:site_name\" content=\"i10X News\" \/>\n<meta property=\"article:published_time\" content=\"2026-11-19T07:29:44+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/i10xnews.kinsta.cloud\/wp-content\/uploads\/2026\/11\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents-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=\"8 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"NewsArticle\",\"@id\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents\"},\"author\":{\"name\":\"Christopher Ort\",\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/#\\\/schema\\\/person\\\/d410bb7ee72febefca2c9c369939219a\"},\"headline\":\"FrontierBreach: Hidden Instructions Hijacked AI Agents in 34.9% of Enterprise Tests\",\"datePublished\":\"2026-11-19T07:29:44+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents\"},\"wordCount\":1691,\"publisher\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/11\\\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents-title.webp\",\"articleSection\":[\"Agents\",\"AI Security\",\"Benchmarks &amp; Evals\",\"Studies\"],\"inLanguage\":\"en-US\",\"dateModified\":\"2026-11-19T07:29:44+00:00\"},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents\",\"url\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents\",\"name\":\"FrontierBreach: Prompt Injection Benchmark Results - i10X News\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/i10x.ai\\\/news\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/11\\\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents-title.webp\",\"datePublished\":\"2026-11-19T07:29:44+00:00\",\"description\":\"FrontierBreach by PromptArmor hid prompt injections in emails, files and skills: attacks worked in 34.9% of enterprise agent tests. Results and caveats.\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents#breadcrumb\"},\"inLanguage\":\"en-US\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"en-US\",\"@id\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents#primaryimage\",\"url\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/11\\\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents-title.webp\",\"contentUrl\":\"https:\\\/\\\/i10x.ai\\\/news\\\/wp-content\\\/uploads\\\/2026\\\/11\\\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents-title.webp\",\"width\":1344,\"height\":768,\"caption\":\"Image: AI-generated, i10X\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/i10xnews.kinsta.cloud\\\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/i10x.ai\\\/news\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"FrontierBreach: Hidden Instructions Hijacked AI Agents in 34.9% of Enterprise Tests\"}]},{\"@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":"FrontierBreach: Prompt Injection Benchmark Results - i10X News","description":"FrontierBreach by PromptArmor hid prompt injections in emails, files and skills: attacks worked in 34.9% of enterprise agent tests. Results and caveats.","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\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents","og_locale":"en_US","og_type":"article","og_title":"FrontierBreach: Prompt Injection Benchmark Results - i10X News","og_description":"FrontierBreach by PromptArmor hid prompt injections in emails, files and skills: attacks worked in 34.9% of enterprise agent tests. Results and caveats.","og_url":"https:\/\/i10xnews.kinsta.cloud\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents","og_site_name":"i10X News","article_published_time":"2026-11-19T07:29:44+00:00","og_image":[{"width":1344,"height":768,"url":"https:\/\/i10xnews.kinsta.cloud\/wp-content\/uploads\/2026\/11\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents-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":"8 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"NewsArticle","@id":"https:\/\/i10xnews.kinsta.cloud\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents#article","isPartOf":{"@id":"https:\/\/i10xnews.kinsta.cloud\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents"},"author":{"name":"Christopher Ort","@id":"https:\/\/i10x.ai\/news\/#\/schema\/person\/d410bb7ee72febefca2c9c369939219a"},"headline":"FrontierBreach: Hidden Instructions Hijacked AI Agents in 34.9% of Enterprise Tests","datePublished":"2026-11-19T07:29:44+00:00","mainEntityOfPage":{"@id":"https:\/\/i10xnews.kinsta.cloud\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents"},"wordCount":1691,"publisher":{"@id":"https:\/\/i10x.ai\/news\/#organization"},"image":{"@id":"https:\/\/i10xnews.kinsta.cloud\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents#primaryimage"},"thumbnailUrl":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/11\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents-title.webp","articleSection":["Agents","AI Security","Benchmarks &amp; Evals","Studies"],"inLanguage":"en-US","dateModified":"2026-11-19T07:29:44+00:00"},{"@type":"WebPage","@id":"https:\/\/i10xnews.kinsta.cloud\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents","url":"https:\/\/i10xnews.kinsta.cloud\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents","name":"FrontierBreach: Prompt Injection Benchmark Results - i10X News","isPartOf":{"@id":"https:\/\/i10x.ai\/news\/#website"},"primaryImageOfPage":{"@id":"https:\/\/i10xnews.kinsta.cloud\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents#primaryimage"},"image":{"@id":"https:\/\/i10xnews.kinsta.cloud\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents#primaryimage"},"thumbnailUrl":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/11\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents-title.webp","datePublished":"2026-11-19T07:29:44+00:00","description":"FrontierBreach by PromptArmor hid prompt injections in emails, files and skills: attacks worked in 34.9% of enterprise agent tests. Results and caveats.","breadcrumb":{"@id":"https:\/\/i10xnews.kinsta.cloud\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents#breadcrumb"},"inLanguage":"en-US","potentialAction":[{"@type":"ReadAction","target":["https:\/\/i10xnews.kinsta.cloud\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents"]}]},{"@type":"ImageObject","inLanguage":"en-US","@id":"https:\/\/i10xnews.kinsta.cloud\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents#primaryimage","url":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/11\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents-title.webp","contentUrl":"https:\/\/i10x.ai\/news\/wp-content\/uploads\/2026\/11\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents-title.webp","width":1344,"height":768,"caption":"Image: AI-generated, i10X"},{"@type":"BreadcrumbList","@id":"https:\/\/i10xnews.kinsta.cloud\/frontierbreach-promptarmor-prompt-injection-benchmark-enterprise-agents#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/i10x.ai\/news"},{"@type":"ListItem","position":2,"name":"FrontierBreach: Hidden Instructions Hijacked AI Agents in 34.9% of Enterprise Tests"}]},{"@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\/9568","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=9568"}],"version-history":[{"count":0,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/posts\/9568\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/media\/9567"}],"wp:attachment":[{"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/media?parent=9568"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/categories?post=9568"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/tags?post=9568"},{"taxonomy":"company","embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/companies?post=9568"},{"taxonomy":"product","embeddable":true,"href":"https:\/\/i10x.ai\/news\/wp-json\/wp\/v2\/products?post=9568"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}