Tag Archives: AI

About Authentication Bypass - Microsoft SharePoint (CVE-2026-55040) vulnerability

About Authentication Bypass - Microsoft SharePoint (CVE-2026-55040) vulnerability

About Authentication Bypass - Microsoft SharePoint (CVE-2026-55040) vulnerability. SharePoint is a web application for corporate intranet portals, document management, and collaboration. This vulnerability was disclosed as part of the July Microsoft Patch Tuesday. A remote unauthenticated attacker can leverage this vulnerability to bypass authentication on a vulnerable SharePoint server and perform operations as a SharePoint site user or administrator. The vulnerability is due to several issues in the JWT token validation pipeline. A JWT (JSON Web Token) is essentially an "electronic pass" that SharePoint uses to determine who a user is and what actions they are authorized to perform. When combined with CVE-2026-63520, the vulnerability allows an attacker to perform remote code execution on a SharePoint server without authentication, with the privileges of the SharePoint service account. As a result, an attacker can gain access to documents and other organizational data stored in SharePoint, modify or delete them, extract credentials, install malware, and establish persistence in the system, and then use the compromised server for further penetration into the organization's internal network.

🛠 The vulnerability was discovered by Stephen Fewer from Rapid7 during SharePoint research conducted in January and March 2026. The research aimed to develop an exploit for the Pwn2Own Berlin competition. AI tools and agentic workflows were actively used throughout the research. Rapid7 published a detailed analysis of the vulnerability and the exploit code on its blog on August 11.

👾 According to VulnCheck, exploitation attempts were detected on honeypots shortly after the exploit was published. The vulnerability was added to the VulnCheck KEV on August 12 and to the CISA KEV on August 18.

⚙️ Updates are available for Microsoft SharePoint Server 2016 (Enterprise), 2019, and Subscription Edition. Note that you need to install the latest SharePoint cumulative updates that include fixes for CVE-2026-55040 and CVE-2026-63520.

🌐 As of August 25, 2026, Censys experts reported around 21,000 on-premises SharePoint installations exposing their version information. According to The Shadowserver Foundation, around 8,500 SharePoint servers were accessible from the Internet each day in September 2026. These are observed servers, and not all of them are necessarily vulnerable. No statistics on vulnerable servers are available.

About Remote Code Execution - NVIDIA Container Toolkit (CVE-2024-0132) vulnerability

About Remote Code Execution - NVIDIA Container Toolkit (CVE-2024-0132) vulnerability

About Remote Code Execution - NVIDIA Container Toolkit (CVE-2024-0132) vulnerability. NVIDIA's bulletin was released on September 25. The vulnerability was found by researchers from Wiz.

Container Toolkit provides containerized AI applications with access to GPU resources. AI is now almost impossible without the use of video cards. 😏 Therefore, this component is very common.

The essence of the vulnerability is that a launched malicious container image can gain access to the host file system, which, in turn, can lead to the attacker's code execution, denial of service, escalation of privileges, information disclosure, and data tampering.

If an attacker gains access to a desktop in this way, it’s not so bad, but what if he gains access to Kubernetes nodes or a cluster? 🫣 AI service providers (a la Hugging Face) that launch untrusted images are at risk.

На русском

Qualys announced the TotalAI module for artificial intelligence (AI) and large language models (LLM) security

Qualys announced the TotalAI module for artificial intelligence (AI) and large language models (LLM) security

Qualys announced the TotalAI module for artificial intelligence (AI) and large language models (LLM) security. The module will be available in Q4 2024 as part of the Enterprise TruRisk platform.

Announced features:

🔹 Detection and monitoring of the AI ​​infrastructure of organizations. To avoid "shadow LLM".

🔹 Vulnerability Management with a focus on AI threats. Especially on countering theft (extraction) of data and models. They will offer a variety of ways to fix vulnerabilities.

🔹 Specialized LLM scanning focussed on prompt injection, model theft, and disclosure of confidential information.

🔹 Compliance Management and risk management. They emphasize combating data leaks and mention GDPR, PCI, CCPA.

There is a screenshot of the interface with statistics on models and related threats. We can also see statistics on threats related to assets and interesting informers for AI Workloads, AI Software and GPU.

На русском