{"id":2768,"date":"2025-07-08T10:00:42","date_gmt":"2025-07-08T01:00:42","guid":{"rendered":"https:\/\/bestpathresearch.com\/?p=2768"},"modified":"2025-07-08T10:31:55","modified_gmt":"2025-07-08T01:31:55","slug":"20250707","status":"publish","type":"post","link":"https:\/\/bestpathresearch.com\/en\/2025\/07\/08\/20250707\/","title":{"rendered":"AI-Assisted Penetration Testing"},"content":{"rendered":"\t\t<div data-elementor-type=\"wp-post\" data-elementor-id=\"2768\" class=\"elementor elementor-2768\" data-elementor-post-type=\"post\">\n\t\t\t\t\t\t<section class=\"elementor-section elementor-top-section elementor-element elementor-element-69825c2 elementor-section-boxed elementor-section-height-default elementor-section-height-default\" data-id=\"69825c2\" data-element_type=\"section\" data-e-type=\"section\">\n\t\t\t\t\t\t<div class=\"elementor-container elementor-column-gap-default\">\n\t\t\t\t\t<div class=\"elementor-column elementor-col-100 elementor-top-column elementor-element elementor-element-30fee80\" data-id=\"30fee80\" data-element_type=\"column\" data-e-type=\"column\">\n\t\t\t<div class=\"elementor-widget-wrap elementor-element-populated\">\n\t\t\t\t\t\t<div class=\"elementor-element elementor-element-afe44be elementor-widget elementor-widget-text-editor\" data-id=\"afe44be\" data-element_type=\"widget\" data-e-type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t\t\tAt Best Path Research, we recently completed a black box web application penetration test for a major Japanese company. By combining manual testing with AI-driven tools, we identified critical vulnerabilities more quickly and effectively than we could have done with manual methods alone. We were able to successfully mitigate a number of serious potential vulnerabilities before the site was released publicly.\n\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-eed5245 elementor-widget elementor-widget-heading\" data-id=\"eed5245\" data-element_type=\"widget\" data-e-type=\"widget\" data-widget_type=\"heading.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<h2 class=\"elementor-heading-title elementor-size-default\">Reinforcement Learning use in Penetration Testing<\/h2>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-f66597c elementor-widget elementor-widget-text-editor\" data-id=\"f66597c\" data-element_type=\"widget\" data-e-type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t\t\tBased on modern generative AI (genAI) and machine learning (ML), we\u2019re now applying reinforcement learning\u00a0(RL) techniques to further automate parts of the web application testing process, starting with popular, and potentially high-risk, open-source blogging and website software. These RL-based agents learn to autonomously identify and probe common security flaws and mis-configurations in an isolated environment, mimicking the ways in which an experienced, malicious, human hacker might attempt to discover and exploit the same vulnerabilities.\u00a0\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-6c4be57 elementor-widget elementor-widget-heading\" data-id=\"6c4be57\" data-element_type=\"widget\" data-e-type=\"widget\" data-widget_type=\"heading.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<h2 class=\"elementor-heading-title elementor-size-default\">Mitigating LLM Exploits in Public-Facing Systems<\/h2>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-1e30630 elementor-widget elementor-widget-text-editor\" data-id=\"1e30630\" data-element_type=\"widget\" data-e-type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t\t\tWe\u2019re also focusing on securing public-facing large language models (LLMs) which can be susceptible to exploits through prompt injection or jailbreak techniques which can leak sensitive information or execute unintended actions. Such exploits are conceptually similar to classic SQL injection vulnerabilities which were a focus in our recent web application pen-testing work. Our future work includes testing frameworks and safeguards to help companies safely deploy LLMs in production.\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-9102f3f elementor-widget elementor-widget-heading\" data-id=\"9102f3f\" data-element_type=\"widget\" data-e-type=\"widget\" data-widget_type=\"heading.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t<h2 class=\"elementor-heading-title elementor-size-default\">Generative AI Threats<\/h2>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-1e41b5e elementor-widget elementor-widget-text-editor\" data-id=\"1e41b5e\" data-element_type=\"widget\" data-e-type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t\t\tThe flip side to the many positive uses of genAI is that they are already being used by attackers to craft convincing phishing emails and realistic social engineering attacks. Recent hacks even use audio and video genAI models to trick employees into revealing sensitive information during telephone calls or in video messages. With our decades of combined experience in AI and ML, we\u2019re also developing tools to detect and defend against these evolving threats before they reach end users.\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<div class=\"elementor-element elementor-element-ef4ae5b elementor-widget elementor-widget-text-editor\" data-id=\"ef4ae5b\" data-element_type=\"widget\" data-e-type=\"widget\" data-widget_type=\"text-editor.default\">\n\t\t\t\t<div class=\"elementor-widget-container\">\n\t\t\t\t\t\t\t\t\tAt Best Path Research, we\u2019re using AI to strengthen both offensive and defensive security capabilities, helping companies stay safe and secure in a rapidly changing threat landscape.\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t<\/section>\n\t\t\t\t<\/div>\n\t\t","protected":false},"excerpt":{"rendered":"<p>At Best P [&hellip;]<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_locale":"en_US","_original_post":"https:\/\/bestpathresearch.com\/?p=2764","footnotes":""},"categories":[27],"tags":[],"class_list":["post-2768","post","type-post","status-publish","format-standard","hentry","category-news_en","en-US"],"_links":{"self":[{"href":"https:\/\/bestpathresearch.com\/wp-json\/wp\/v2\/posts\/2768","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/bestpathresearch.com\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/bestpathresearch.com\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/bestpathresearch.com\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/bestpathresearch.com\/wp-json\/wp\/v2\/comments?post=2768"}],"version-history":[{"count":6,"href":"https:\/\/bestpathresearch.com\/wp-json\/wp\/v2\/posts\/2768\/revisions"}],"predecessor-version":[{"id":2775,"href":"https:\/\/bestpathresearch.com\/wp-json\/wp\/v2\/posts\/2768\/revisions\/2775"}],"wp:attachment":[{"href":"https:\/\/bestpathresearch.com\/wp-json\/wp\/v2\/media?parent=2768"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/bestpathresearch.com\/wp-json\/wp\/v2\/categories?post=2768"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/bestpathresearch.com\/wp-json\/wp\/v2\/tags?post=2768"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}