<\/span><\/h3>\nWhile the AI agent performed the repetitive and systematic tasks of reconnaissance, the human operator was crucial in providing the strategic direction and initial access. Hunt.io recovered a hidden web shell, a piece of malicious code designed to provide remote access, planted on a ministry web server. This indicates that the operator was already inside the network before deploying the Hermes agent. Additionally, the investigation uncovered scripts specifically written to interact with internal Hadoop systems and a mail-testing script containing hardcoded stolen mailbox passwords.<\/p>\nThe recovered data does not yet show any evidence of data exfiltration from the ministry’s network, and the exact method by which the operator initially gained access remains unknown. Thailand’s national Computer Emergency Response Team (CERT) and cybersecurity agency were notified of the breach on July 15, 2024. As of July 24, 2024, neither agency had released any public statements regarding the incident.<\/p>\n
The operational scripts revealed a crucial detail: they were designed to exploit a known vulnerability in Hadoop database services that, by default, accept any password. This allowed the operator to gain unauthorized access to the vast data stores managed by the ministry. The human element was also evident in the sophisticated targeting. Hunt.io’s analysis showed that the password list used by the operator was not a generic dictionary but was constructed using the ministry’s own departmental abbreviations. Furthermore, shellcode was found to contain hardcoded paths pointing into the ministry’s internal network, demonstrating a deep understanding of the target environment.<\/p>\n
<\/span>The Autonomous Agent’s Actions: A Step-by-Step Breakdown<\/span><\/h3>\nThe Hermes agent’s activities are meticulously documented in five files named call_00_*.txt<\/code>. These logs provide a chronological account of the agent’s actions within the ministry’s network. The sequence of operations included:<\/p>\n\nKernel Vulnerability Scanning:<\/strong> The agent initiated scans against a ministry host, searching for potential vulnerabilities in the Linux kernel.<\/li>\nSecond LinPEAS Run:<\/strong> A subsequent execution of LinPEAS, a well-known script designed to identify privilege escalation pathways on Linux systems.<\/li>\nElevated-Permission Binary Sweep:<\/strong> The agent searched for files with elevated permissions, a common tactic to identify potential weak points for privilege escalation.<\/li>\nFilesystem Listing:<\/strong> A comprehensive listing of files and directories within accessible parts of the file system.<\/li>\nRecursive Directory Crawl:<\/strong> The agent performed a deep, recursive crawl of the web root belonging to the Office of the Permanent Secretary.<\/li>\n<\/ul>\nThis crawl uncovered a significant volume of sensitive documents, including Office documents, performance evaluations, and, critically, personnel records dating back to 2012. While the logs confirm the agent’s access to and reading of these directories, they do not indicate that any of the files were actually exfiltrated.<\/p>\n
Exploiting Kernel Vulnerabilities:<\/strong> The customized LinPEAS script handed to the agent was not a standard version. It specifically checked for four known Linux kernel vulnerabilities: Copy Fail (CVE-2026-31431), Dirty Frag (CVE-2026-43284 and CVE-2026-43500), and DirtyClone (CVE-2026-43503). These vulnerabilities, when successfully exploited under specific conditions, can grant a local user root privileges. However, the recovered materials do not specify the ministry’s kernel version, nor do they confirm whether any of these exploits were successfully executed. The operator’s scripts were prepared weeks in advance of the attack, suggesting a well-planned operation.<\/p>\nAttribution Indicators:<\/strong> Further analysis of the attack infrastructure provides potential clues about the perpetrators. The operator’s SSH session into the staging server originated from an IP address associated with Hong Kong (103.97.0[.]57). The web interface password for the Hermes agent contained the Chinese word "Leishen," meaning "thunder god." Additionally, an API key for FOFA, a Chinese asset-search service, was found alongside this password. The same staging server had previously hosted a ShadowPad controller and currently runs a VShell command-and-control listener. Based on these indicators, Hunt.io assesses with low-to-medium confidence that the operator is Chinese-speaking or fluent in the language. However, no specific threat group has been definitively linked to this attack.<\/p>\n<\/span>The Vulnerable Gateway: Hadoop and Unsecured Services<\/span><\/h3>\nA significant portion of the custom code deployed by the attacker was directed at the ministry’s Hadoop cluster, a system designed for storing and querying massive datasets. A script named hive_rce_py2.py<\/code> was observed connecting to HiveServer2, the SQL interface for the Hadoop cluster, on an internal machine at port 10000. The connection was established by sending a password.<\/p>\nThis exploit hinges on a critical security lapse: Apache’s own documentation for HiveServer2 indicates that the default authentication mode is "NONE," meaning it accepts any provided password without verification. This configuration flaw effectively turned the SQL front-end into an open door for unauthorized access.<\/p>\n
Once connected, the script installed a malicious Java add-on, HiveCmd.jar<\/code>, as a user-defined function (UDF). UDFs allow users to execute operating system commands through ordinary database queries and retrieve the results. Cloudera, a prominent Hadoop vendor, has warned that the ability to install such functions can enable attackers to run arbitrary code with the privileges of the Hive service account, granting them access to highly sensitive data.<\/p>\nAdditional Exploitation Attempts:<\/strong> Beyond the Hadoop cluster, the attacker also deployed a previously undocumented Go implant named Hades. This implant was compiled for both Windows and Linux, with 62 copies found on the staging server. While Hunt.io analyzed one copy of each operating system, the remaining 60 were not individually examined. The hardcoded addresses within Hades suggest a connection to a second Hong Kong-based host, though no recovered artifact confirms Hades successfully communicating with a ministry machine.<\/p>\nFurthermore, separate scripts were found that tested default credentials against an internal GlassFish console. Additionally, exploit code for three older vulnerabilities affecting polkit<\/code>, sudo<\/code>, and IIS 6.0 was present, indicating a broad reconnaissance effort across various potential attack vectors.<\/p>\n<\/span>The Unseen Trail: How the Attack Was Uncovered<\/span><\/h3>\nThe method by which the attack was ultimately revealed is as critical as the attack itself. The Hermes agent, while designed for autonomy, still leaves a digital footprint. The agent’s web panel typically returns a "HermesWebUI" server header. A search for this header, as of Hunt.io’s July 23 report, revealed approximately 5,900 scan events over a month, indicating numerous instances of the agent being used.<\/p>\n
However, the more significant discovery point was the consistent presence of a \/hermes-results\/<\/code> folder where the agent reliably stored its findings. This folder, with its predictable naming conventions for the files within, was indexed by Hunt.io’s tools. The search turned up 575 hits on July 23, each representing a host and filename pair. Crucially, it was not a sophisticated security control that exposed the operator, but rather a simple directory listing enabled on the web server. This highlights how basic misconfigurations can have catastrophic security implications.<\/p>\nThe implications of this attack are far-reaching. The seamless integration of AI into the cyberattack lifecycle blurs the lines between human-led operations and autonomous malicious activity. The "YOLO" mode employed by Hermes effectively removes the human element from the immediate execution loop, allowing for rapid and relentless probing of target systems. While the human operator still directs the overall campaign, the AI agent can operate tirelessly, identifying and exploiting vulnerabilities at machine speed.<\/p>\n<\/span>Broader Impact and Implications<\/span><\/h3>\nThe Ministry of Finance incident serves as a stark warning about the evolving threat landscape in the age of artificial intelligence. The ability to deploy AI agents that can operate with minimal human oversight presents a significant challenge for cybersecurity professionals. Traditional security measures, often designed to detect human-like behavior or specific attack patterns, may struggle to keep pace with the adaptability and speed of AI-driven attacks.<\/p>\n
The Democratization of Sophisticated Attacks:<\/strong> Open-source AI tools like Hermes, when misused, can lower the barrier to entry for sophisticated cyberattacks. Individuals or groups with limited technical expertise can leverage these tools to conduct complex operations that were previously the domain of highly skilled nation-state actors.<\/p>\nThe Need for Enhanced AI Security:<\/strong> This incident underscores the urgent need for developers of AI tools to implement robust security features and responsible deployment guidelines. While Hermes itself is not a hacking tool, its features can be exploited. The "YOLO" mode, while useful for legitimate automation, becomes a dangerous weapon when unchecked.<\/p>\nProactive Threat Hunting and Incident Response:<\/strong> The discovery by Hunt.io and Bob Diachenko highlights the importance of proactive threat hunting and robust incident response capabilities. The ability to quickly analyze attacker infrastructure and identify exposed data is critical in mitigating the impact of such breaches.<\/p>\nGovernment and Critical Infrastructure Vulnerabilities:<\/strong> The targeting of a national Ministry of Finance emphasizes the vulnerability of critical government infrastructure to advanced cyber threats. The compromise of such an entity could have profound implications for national security and economic stability.<\/p>\nFuture Outlook:<\/strong> As AI technology continues to advance, it is inevitable that both offensive and defensive cybersecurity capabilities will evolve in tandem. This incident is likely just the beginning of a new era of AI-driven cyber warfare, demanding continuous innovation and vigilance from the global cybersecurity community. The key takeaway is that the combination of human strategic intent and autonomous AI execution creates a potent and formidable threat that requires a fundamental rethinking of our defense strategies.<\/p>\n","protected":false},"excerpt":{"rendered":"An sophisticated cyberattack campaign has successfully infiltrated the network of Thailand’s Ministry of Finance, exploiting a popular open-source AI assistant and leveraging misconfigurations to probe for sensitive data. The operation, uncovered by threat intelligence firm Hunt.io and researcher Bob Diachenko, highlights a concerning evolution in AI-assisted cyber warfare, where autonomous agents are increasingly being deployed …<\/p>\n","protected":false},"author":27,"featured_media":6949,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[108],"tags":[734,109,258,3411,173,1579,111,110,3441,3440],"class_list":["post-6950","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-cybersecurity-protection","tag-campaign","tag-cybersecurity","tag-driven","tag-espionage","tag-finance","tag-ministry","tag-privacy","tag-security","tag-targeted","tag-thailand"],"_links":{"self":[{"href":"https:\/\/lockitsoft.com\/index.php?rest_route=\/wp\/v2\/posts\/6950","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/lockitsoft.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/lockitsoft.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/lockitsoft.com\/index.php?rest_route=\/wp\/v2\/users\/27"}],"replies":[{"embeddable":true,"href":"https:\/\/lockitsoft.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=6950"}],"version-history":[{"count":0,"href":"https:\/\/lockitsoft.com\/index.php?rest_route=\/wp\/v2\/posts\/6950\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/lockitsoft.com\/index.php?rest_route=\/wp\/v2\/media\/6949"}],"wp:attachment":[{"href":"https:\/\/lockitsoft.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=6950"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/lockitsoft.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=6950"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/lockitsoft.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=6950"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}