This vulnerability is a “Route Confusion and SQL Injection” vulnerability found in WordPress. An attacker may attempt to carry out an attack by sending a JSON request containing the syntax of a malicious SQL injection attack to the /wp-json/batch/v1 endpoint, exploiting incorrect sub-request handling logic. The security patch for this vulnerability was released on July 17, 2026, and the AIWAF product detects it using SQL injection-related patterns. — 1. Overview WordPress is an open-source content management system (CMS) that powers a significant portion of the world’s websites. With its vast ecosystem of themes and plugins, as well as its ease of customization, it helps users quickly build and operate a wide variety of sites, ranging from blogs to corporate websites. This report summarizes the analysis of the WP2Shell vulnerabilities—CVE-2026-63030 and CVE-2026-60137—that were discovered in this system. — 2. Attack Type Dubbed “WP2Shell,” this vulnerability is a Remote Code Execution (RCE) vulnerability found in the default WordPress environment. It involves two interlinked CVE vulnerabilities that work together to execute malicious commands. CVE-2026-63030 is a path confusion vulnerability in the REST API of the default WordPress environment. The /wp-json/batch/v1 endpoint in WordPress can accept JSON-formatted REST sub-requests for validation and execution-related REST sub-requests in JSON format. If an error occurs in one sub-request, the handler processing that request shifts the values within the array where those requests are stored; as a result, the subsequent sub-request is executed within the handler and authorized context of the previous sub-request. An attacker can exploit this by inserting a sub-request designed to trigger an error, thereby causing the subsequent sub-request to execute without validation. Attack Request: POST /wp-json/batch/v1 HTTP/1.1 Host: www.test.com Content-Type: application/json Accept: application/json { “validation”: “normal”, “requests”: [ { “path”: “/wp/v2/posts”, “method”: “GET” }, { “path”: “://invalid-url”, “method”: “GET” }, { “path”: “/wp/v2/posts”, “method”: “GET” } ] } CVE-2026-60137 is an SQL injection vulnerability that occurs in the default WordPress environment. An attacker can exploit this vulnerability by sending and executing a malicious SQL query through the `author_exclude` parameter exposed in the REST API. However, this vulnerability previously required authentication to access the endpoint supporting this parameter; this requirement can be bypassed by exploiting the CVE-2026-63030 vulnerability. Final Attack Request: POST /wp-json/batch/v1 HTTP/1.1 Host: www.test.com Content-Type: Read more about [2026.07 Vulnerability Report] WP2Shell (CVE-2026-63030, CVE-2026-60137)[…]

Oracle PeopleSoft PeopleTools is a widely deployed Enterprise Resource Planning (ERP) platform used by universities, hospitals, and government agencies worldwide to manage HR, payroll, and student records. CVE-2026-35273 is a critical zero-day vulnerability in the PeopleTools Environment Management Hub (PSEMHUB) component, allowing an unauthenticated remote attacker to achieve full OS-level code execution by chaining a Read more about [2026.06 Vulnerability Report] Oracle PeopleSoft PeopleTools RCE (CVE-2026-35273)[…]

This vulnerability is a ‘PHP Object Injection’ vulnerability discovered in Adobe Commerce (Magento). Attackers exploit the fact that the ServiceInputProcessor component attempts deserialization without validation or inspection to send serialized malicious PHP objects to vulnerable endpoints for execution. A security patch for this vulnerability was released in September 2025, and our AIWAF products plan to Read more about [2026.06 Vulnerability Report] SessionReaper: Adobe Commerce PHP Object Injection[…]

CVE‑2026‑49975, known as “HTTP/2 Bomb,” is a Denial of Service (DoS) vulnerability that leads to severe server overload during HTTP/2 request processing. Attackers can exploit the intersection of HPACK header compression and HTTP/2 flow control to trigger massive memory expansion with minimal network traffic. If successfully exploited, this vulnerability can cause a rapid spike in Read more about [2026.06 Vulnerability Report] HTTP/2 Bomb Server Overload DoS (CVE-2026-49975)[…]

Palo Alto Networks PAN-OS firewalls are critical infrastructure assets protecting enterprise network perimeters. CVE-2026-0300 is a critical zero-day buffer overflow vulnerability in the User-ID™ Authentication Portal (Captive Portal) service that allows an unauthenticated, remote attacker to execute arbitrary code with root privileges by sending specially crafted packets. With confirmed in-the-wild exploitation by advanced threat actors Read more about 2026.05 – PAN-OS User-ID Authentication Portal Buffer Overflow Vulnerability (CVE-2026-0300)[…]

This vulnerability is a ‘Remote Code Execution’ vulnerability discovered in Apache ActiveMQ, where an attacker constructs an attacker server address that returns malicious XML code using the VM transport method and sends it to the Apache ActiveMQ /api/jolokia endpoint for execution. A security patch for this vulnerability was released on March 30, 2026, and AIWAF Read more about 2026.05 – Apache ActiveMQ RCE (CVE-2026-34197)[…]

CVE-2026-7482 is a high-risk vulnerability in Ollama that occurs during GGUF model file processing. The issue is caused by a heap-based out-of-bounds read, which may allow unintended memory data from the Ollama server process to be exposed. If exploited, an attacker could submit a crafted GGUF model file and potentially leak sensitive information such as API keys, environment variables,  system prompts, internal instructions, and user conversation data. The risk is higher when Ollama APIs such as /api/create and /api/push are exposed without proper access control. This vulnerability affects Ollama versions prior to 0.17.1. Affected environments should upgrade to Ollama 0.17.1 or later, restrict external API access, and review exposed instances for suspicious model creation or data transfer activity. 1. Overview CVE-2026-7482 is a high-severity vulnerability discovered in Ollama, an open-source platform that enables users to run Large Language Models (LLMs) locally and in server environments. This flaw is classified as a Heap-based Out-of-Bounds Read, which occurs during the process of loading and parsing GGUF model files within Ollama. Attack Vector & Impact An attacker can trigger this vulnerability by uploading a specially crafted, malicious GGUF file to a vulnerable Ollama server. Successful exploitation allows unauthorized reading of the server process memory space. This can lead to the exposure of highly sensitive information, including environment variables, integrated API keys, system prompts, and chat histories of other users. Exposure Risks In certain Ollama deployment environments, APIs associated with model creation and uploading (such as /api/create and /api/push) may be exposed to the internet without proper authentication. In such scenarios, an attacker can cause severe data leakage simply by submitting a manipulated model file. Affected Versions & Mitigation This vulnerability affects all Ollama versions prior to 0.17.1. Vendors and security advisories strongly recommend that users immediately upgrade to version 0.17.1 or higher, where the patch has been applied. 2. Root Cause Analysis and Attack Vector Attack Flow and Accessibility Initial Access: CVE-2026-7482 can be exploited via exposed Ollama APIs without requiring any prior authentication. Remote attackers can easily access the vulnerability if the Ollama server is exposed to an external network or bound to 0.0.0.0. Exploitation Process: The attack begins when a perpetrator submits a malicious GGUF file containing manipulated metadata to the server. The attacker then requests model creation based on this file, triggering an out-of-bounds memory read within the server process. Exfiltration Mechanism: The leaked memory data is captured and embedded into the newly generated model artifact. The attacker can then exploit Ollama’s model registry push feature to exfiltrate this sensitive data to an external server or a registry under their control. Root Cause The core vulnerability stems from insufficient validation of GGUF file metadata and inadequate memory boundary checks. Ollama fails to cross-verify the tensor offset and size information declared in the GGUF file against the actual file size. Consequently, this allows the system to attempt to read data outside the file’s legitimate boundaries. When the model processing and quantization logic references the buffer—blindly trusting the declared metadata—it reads beyond the boundaries of the allocated heap buffer into other memory spaces within the server process. This results in a Heap-based Out-of-Bounds Read. Technical analyses indicate that this flaw specifically occurs within the WriteTo() function during the GGUF loading and quantization processes. Exfiltrated Data Targets Server memory data exposed through the out-of bounds read is incorporated into the resulting model artifacts. By pushing these artifacts to an external registry, attackers can exfiltrate highly sensitive information. Publicly available analyses highlight that the following types of data are at risk of exposure: System Control Information Critical environment variables, including API keys, authentication tokens, and database connection strings. AI Model Assets System prompts, internal instructions, and security policy configurations. User Data Prompts, chat histories, and outputs from external tool integrations belonging to other users. 3. Impact and Business Implications Threat Scale and Attack Indicators Global Exposure: Recent security research indicates that a significant number of Ollama servers worldwide are currently exposed to the internet. High-Severity Risk: CVE-2026-7482 is a Critical-rated vulnerability with a CVSS score of 9.1. Because it can be exploited without authentication, the risk is exceptionally high for any publicly exposed environment. Exploitation Potential: With Proof-of-Concept (PoC) exploits publicly available, there is a severe risk of automated scanning and large-scale, opportunistic attacks. Core Asset and Sensitive Data Leakage A successful exploitation allows attackers to exfiltrate critical information residing in the server’s memory. This includes: Environment Variables: API keys, authentication tokens, and database connection strings. Read more about [2026.05 Vulnerability Report] Ollama Memory Information Disclosure Vulnerability Analysis (CVE-2026-7482)[…]

Anthropic Mythos (Claude Mythos Preview) is a next-generation AI model designed to go beyond standard conversational AI, specifically optimized for software source code analysis, vulnerability detection, and attack scenario prediction. It demonstrates the capability to identify high-risk zero-day vulnerabilities in major operating systems and web browsers—tasks that previously required extensive manual effort from highly skilled Read more about [2026.04 Vulnerability Report] Anthropic Mythos[…]

This report provides a technical analysis of the publicly available KillChain project (oxfemale/KillChain), focusing on how the tool exploits CVE-2026-0828 in Safetica’s vulnerable DLP kernel driver (ProcessMonitorDriver.sys) to forcibly terminate protected Windows processes (including PPL) and security-related processes. The scope of analysis includes KillChain’s internal execution flow, vulnerable driver loading, invocation of the kernel interface Read more about [2026.04 Vulnerability Report] BYOVD in Practice: KillChain and CVE-2026-0828 Analysis[…]

Cisco Integrated Management Controller (IMC) is a critical infrastructure asset that provides out-of-band management for UCS servers. CVE-2026-20093 is a critical authentication bypass vulnerability that allows a remote, unauthenticated attacker to gain full administrative access by exploiting a flaw in the password reset logic. With proof-of-concept (PoC) scripts being distributed in the dark web and Read more about [2026.04 Vulnerability Report] Cisco IMC Authentication Bypass Vulnerability (CVE-2026-20093)[…]