November 2025 Summaries

Quantum computing, once a complex concept, is making strides towards mainstream relevance, presenting both opportunities and challenges, particularly in the realm of cybersecurity. As quantum computers advance, they threaten to render current encryption methods like RSA and Elliptic Curve Cryptography obsolete, prompting a need for post-quantum cryptography standards. Governments and tech companies are actively developing strategies and technologies to address these impending security challenges, with efforts led by organizations like NIST to standardize quantum-resistant algorithms. Cloud service providers such as AWS, Google, and Microsoft are already integrating these algorithms into their systems, while Quantum-as-a-Service (QaaS) accelerates the development of quantum software without requiring access to fragile hardware. The shift towards quantum computing is expected to enhance classical systems rather than replace them, particularly in fields like AI, where quantum's faster processing speeds can improve model training and optimization. As the potential for quantum threats grows, organizations are urged to prepare by assessing their current encryption methods and planning migrations to post-quantum standards, ensuring resilience in the face of these emerging technological shifts.

The new version of the Shai-Hulud worm, which began its spread on November 24, 2025, has impacted over 25,000 GitHub repositories by using compromised NPM packages to steal credentials and exfiltrate them. This iteration, more sophisticated than its predecessor, executes during the pre-installation phase, utilizes a JavaScript runtime called "bun," and introduces new functionalities for persistence and data exfiltration. It leverages cloud modules and creates GitHub repositories to exfiltrate secrets, using a self-hosted GitHub Actions runner as a backdoor. Security teams can detect the worm through suspicious activities associated with NPM install commands. To mitigate the worm's effects, affected users must remove compromised packages, clear NPM caches, rotate exposed credentials, and audit their environments for unauthorized changes. The incident highlights the critical need for runtime threat detection in monitoring third-party packages for malicious activity, with tools like Falco and Sysdig Secure offering necessary visibility and detection capabilities.

Cloud security is facing a critical challenge as rapid development and fragmented tools lead teams to compromise between speed and security. The industry is urged to move beyond "good enough" solutions and embrace the "right way" to secure cloud environments, which involves three core pillars: agentic AI that provides actionable insights rather than merely generating alerts, open innovation that encourages collaboration and transparency over proprietary systems, and runtime insights that offer real-time visibility into live cloud systems. Cloud-Native Application Protection Platforms (CNAPPs) are presented as a unified approach to cloud security, though not all are equally effective. Effective CNAPPs consolidate security functions across the cloud lifecycle, enhancing real-time understanding of threats and vulnerabilities. By adopting these best practices, security can evolve from static visibility to dynamic defense, enabling teams to protect their systems without sacrificing speed or innovation.

An AI Bill of Materials (AIBOM) serves as a comprehensive documentation tool that clarifies the full AI infrastructure within an organization, akin to a Software Bill of Materials (SBOM) for traditional applications. It provides a roadmap for security leaders to identify and assess risks in AI systems, which encompass components like GPUs, containers, datasets, and APIs. As organizations rapidly adopt AI for automating processes and gaining competitive advantages, security teams face the dual challenge of integrating AI into security workflows while simultaneously securing the broader AI implementations across the company. Although AI brings familiar security challenges to new contexts, it relies on existing cloud-native infrastructures, such as containerized base images and orchestration platforms, which security teams already manage. Sysdig's paper on AIBOM emphasizes that understanding AI-specific risks and applying known security practices to AI workloads can demystify AI and ensure secure innovation.

CVE-2024-1086 is a critical vulnerability in the Linux kernel's netfilter component, actively exploited in ransomware campaigns due to its ability to grant root privileges, making it a significant threat to Linux-based systems. Discovered in January 2024 and patched by Linux the following month, this use-after-free vulnerability has existed for over a decade, affecting kernel versions from 3.15 to 6.8-rc1. The flaw arises from improper validation of verdict parameters in the nft_verdict_init() function, leading to a double-free condition that attackers can exploit to execute malicious code and gain root access. This vulnerability is particularly concerning for multi-tenant Linux systems, container hosts, and internet-exposed servers, as it enables post-compromise escalation, defense evasion, and lateral movement. Public proof-of-concept code has made exploitation more accessible, and Sysdig Secure offers detection capabilities to identify potential threats. Given the vulnerability's critical nature and decade-long existence, organizations are urged to prioritize patching to protect their Linux infrastructure from ransomware attacks.

Securing Kubernetes environments presents unique challenges due to their inherent scale, dynamic nature, and complexity, which are exacerbated by a global shortage of skilled security professionals and the increasing sophistication of adversaries. Agentic AI, as an advancement over generative AI, offers a proactive approach to cloud security by autonomously identifying, prioritizing, and aiding in the remediation of issues within Kubernetes and cloud environments. This technology acts as an automated security analyst, capable of triaging vulnerabilities and providing contextual insights to streamline threat investigation and response. Sysdig's implementation of agentic AI, known as Sysdig Sage, enhances security operations by reducing noise, prioritizing vulnerabilities, and accelerating remediation processes, ultimately improving mean time to respond (MTTR) and reducing costs associated with incident response. By integrating AI into workflows, security teams can more effectively manage the defense of cloud-native applications, ensuring quicker, more informed decision-making to mitigate risks and protect sensitive data.

Agentic cloud security represents a transformative approach to cybersecurity, leveraging AI to enhance risk management and enable strategic decision-making for security teams. Traditional AI tools have often fallen short in the fast-paced cloud environment, where threats evolve rapidly, and defenders are inundated with alerts. Agentic AI distinguishes itself by understanding the business context of potential risks, enabling it to make autonomous decisions that prioritize real and exploitable threats while reducing cognitive burdens on security teams. This approach aligns security with business objectives, acting as a critical enabler of innovation and efficiency. Sysdig Sage exemplifies this next-generation AI, offering capabilities that shift security practices from reactive to responsive, thereby empowering organizations to protect critical assets and drive forward-looking business growth.

The Sysdig Threat Research Team (TRT) focuses on developing effective detection rules to identify reverse shells, a common remote access tool used by attackers to control compromised systems. Reverse shells initiate a connection from the victim's machine to the attacker's system, bypassing firewall restrictions on incoming connections. The TRT analyzes various reverse shell techniques, including direct shell execution with network-redirected input/output, indirect shell execution using interprocess communication, and direct command execution without a traditional shell. By understanding the anatomy of reverse shells and their associated system calls, Sysdig TRT creates rules that balance reducing false positives with broad detection capabilities. They continuously refine these rules to improve accuracy and adapt to evolving threats, ensuring robust protection for Sysdig customers and open source Falco users against the latest adversarial techniques in the cloud landscape. These efforts involve implementing new fields in detection rules and leveraging stateful workload detection policies to enhance the relevance and precision of threat detection.

The blog post highlights the integration of Model Context Protocol (MCP) servers with AI tools, particularly focusing on combining Sysdig MCP Server with Amazon Q Developer in VS Code to enhance security workflows. This integration supports a "shift-left" approach in software development, enabling developers to identify and address infrastructure as code (IaC) misconfigurations, vulnerabilities, and potential attack paths during the development phase rather than after deployment. By leveraging these tools, developers can scan IaC files, analyze container images, and predict the security posture of their code, thereby minimizing risks before entering the CI/CD pipeline. The post emphasizes the reduced barrier to integrating diverse technologies with MCPs and encourages exploring additional MCP servers to further improve workflows and security practices.

Falco and Stratoshark have enhanced the capabilities of open-source runtime security by bridging the gap between real-time threat detection and detailed forensic analysis. Falco, a project under the Cloud Native Computing Foundation (CNCF), has introduced features such as automatic forensic capture recording, which generates system capture files during suspicious activities, providing immediate evidence of system behavior. These files can be directly examined in Stratoshark, allowing teams to transition seamlessly from detection to investigation without relying on multiple tools. Additionally, Falco's plugin API now supports field offset mapping, enabling precise tracing of parsed metadata back to raw data, thus enhancing the accuracy of investigations. This integration creates a unified workflow that allows security teams to respond quickly and confidently to threats, maintaining a transparent and efficient open-source framework suitable for modern infrastructure environments. This new model will be demonstrated at KubeCon North America, showcasing its potential to transform security operations by providing comprehensive visibility and real-time insights across containers, hosts, and cloud workloads.

Published by Alejandro Magallon, the blog post discusses the integration of AI agents with GitHub's Model Context Protocol (MCP) server to address security issues in code repositories. It experiments with using OpenAI's Codex and GitHub's MCP server to identify, contextualize, and propose solutions for security vulnerabilities in code, specifically focusing on unauthorized modifications to critical system paths. The article explores how AI agents can assist in managing code repositories, analyzing code without downloading it, and automating issue creation, providing an example of Codex creating a GitHub issue with detailed context and solutions. While highlighting the potential of AI and MCPs to streamline incident response and enhance code security, the article notes the current limitations of AI, including inconsistencies in outputs and vulnerability to manipulation, emphasizing the need for human oversight to ensure accuracy and security.

On November 5, 2025, a researcher from SUSE disclosed three vulnerabilities affecting runc, a container runtime used by platforms like Docker and Kubernetes, which could potentially allow attackers to bypass container isolation and gain root access to the host system. These vulnerabilities involve exploiting race mount conditions and procfs write redirects, requiring the ability to start containers with custom mount configurations, thus making untrusted container images and Dockerfiles likely delivery methods for exploits. The vulnerabilities, identified as CVE-2025-31133, CVE-2025-52565, and CVE-2025-52881, exploit various flaws in runc's handling of maskedPaths and /dev/console mount races, as well as LSM bypass techniques. The Sysdig Threat Research Team has provided recommendations for mitigations and detections, urging users to update to runc versions 1.2.8, 1.3.3, or 1.4.0-rc.3 or later, enable user namespaces, use rootless containers, and apply vendor patches. No active exploits have been identified, but detection tools like Sysdig Secure and Falco can monitor suspicious symlink behaviors to prevent potential exploitation.

The blog post examines the importance of securing Large Language Model (LLM) applications and the role of the OWASP Top 10 framework in mitigating potential vulnerabilities associated with AI workloads. It highlights the evolution and growing adoption of Generative AI (GenAI) technologies, such as Llama 2 and Midjourney, and the consequent need for robust security frameworks, especially in cloud-native environments that utilize platforms like Kubernetes. The text underscores the distinction between AI, machine learning (ML), and LLMs, emphasizing the unique security challenges each presents and the necessity for comprehensive governance and regulatory approaches, such as the EU AI Act. It also discusses the shared responsibility of developers, security teams, and leadership in ensuring trustworthy AI systems and the integration of OWASP Top 10 strategies into existing security practices to reduce exposure to AI-related threats. Additionally, the post advocates for the incorporation of AI components into asset inventories and the use of tools like Sysdig for enhanced risk management and posture reporting.

October 2025 was a significant month for cybersecurity, highlighted by both vulnerabilities and service outages, underscoring the importance of resilience and response speed. Notable incidents included a critical vulnerability in Redis, a widely used open-source data store, which allowed remote code execution and required immediate patching to prevent potential exploitation. Additionally, service disruptions occurred in AWS's Northern Virginia region due to an automation failure and in Azure due to a software defect, both necessitating manual intervention for recovery. These events emphasized the need for distributing services across regions, regularly auditing automation processes, and maintaining redundancy in configuration practices. Other significant developments included the theft of sensitive data from F5 BIG-IP devices and the fallout from a Salesforce supply chain breach, while the international community took a step forward with the signing of a global treaty to combat cybercrime. The month concluded with reflections on the interconnectedness of digital infrastructure and the importance of proactive measures to ensure cybersecurity resilience.