Security Review #278

This is a crash course on the evolution of Machine Learning and modem AI, Large Language Models, and the security implications that come with them.

We demonstrate that we can eliminate the C2 server entirely and create truly autonomous malware, and create an entirely local, C2-less malware that can autonomously discover and exploit one type of privilege escalation vulnerability.

This article presents a proof of concept (PoC) that demonstrates how adversaries can use indirect prompt injection to silently poison the long-term memory of an AI Agent, allowing injected instructions to persist and potentially exfiltrate conversation history.

Singularity is a Loadable Kernel Module (LKM) rootkit developed for Linux 6.x kernels that demonstrates advanced evasion and persistence techniques. This article shows its architecture, from the ftrace-based hooking infrastructure to the anti-forensics mechanisms.

In this article, I will experiment with the technique of injecting code into the protected processes of several Antivirus programs. After a successful injection, I can perform actions that are not permitted for regular processes (regardless of the user permissions). The simplest example is writing a backdoor into the installation folder of the Antivirus.

Acquiring memory on a Windows host is deceptively simple - until you realize the difference between a usable memory dump and a useless blob of bits. In this part we will focus on acquiring three types of memory dumps: full memory capture, process memory dumps, and crash dumps.

In this blog post, we provide an in-depth look at how Adobe Experience Manager (AEM) operates under the hood, including modern dispatcher bypasses that target real systems, and discuss several CVEs identified during our analysis of the AEM source cod

ASCII Smuggling is a technique rooted in the abuse of the Unicode standard, specifically utilizing invisible control characters to embed hidden instruction. In this article, we demonstrate how this technique can be leveraged to abuse LLMs.

We detail a 13-year flaw in Redis (CVE-2025-49844). The vulnerability exploits a Use-After-Free (UAF) memory corruption bug. It allows a post-auth attacker to send a specially crafted malicious Lua script to escape from the Lua sandbox and achieve arbitrary native code execution on the Redis host. This grants an attacker full access to the host system.

MCP security today is fundamentally broken at the architectural level. The combination of overprivileged credentials, no runtime boundaries, and invisible operations creates risk for many organizations, and also the entire ecosystem of open-source tooling. We introduce MCP Snitch, a proxy model with a focus on practical security: allow list-based controls and API key protection.

We investigate spear phishing campaigns that aim to socially engineer targets into clicking links that led to a remotely hosted archive containing a malicious payload. This blog post outlines technical details, and the evidence that led us to assess with a high degree of confidence they employs Large Language Models (LLMs) to assist with their operations.

This blog post presents CVE-2025-3600, an Unsafe Reflection vulnerability in Progress Telerik UI for ASP.NET AJAX. We will demonstrate that, depending on the targeted environment, CVE-2025-3600 can enable Remote Code Execution across a wide range of enterprise-grade solutions.

Because of a relatively insufficient hindsight regarding new “post-quantum” key exchange schemes, most institutions incentivize the use of hybrid schemes, combining the robustness of renowned classical schemes with the post-quantum protection offered by newer schemes. This blogpost details the state of the main concepts of hybrid key exchanges.

Memory forensics shows you how activities unfolded in real time. In this part, we will delve into Volatility, an open-source, cross-platform, modular, and extensible memory analysis tool that is capable of giving you full visibility into what was running when your system got burned.

We detail how to spot 6 common security vulnerabilities in Go code reviews: directory traversal, weak randomness, hostname validation flaws, timing attacks, zip slip, and hardcoded secrets. Practical examples!

This post demonstrates how to inspect the network communication between Defender for Endpoint and its cloud services. In this analysis, I identified several issues, including an authentication bypass, data/command spoofing, information disclosure, and the ability to upload malicious files to investigation packages destined for security analysts.

We explore how Windows implements Pointer Authentication (PAC) on ARM64 -covering bootloader setup, per-process keys, HyperGuard, and memory-safety defenses. We go through the basics, detail exploits mitigation and review the current limitations and future potential.

We detail reassembly attacks, a form of client-side attacks, including HTML smuggling, bypassing conventional security controls to deliver malicious payloads onto endpoints. We also review how attackers leverage advanced technologies like WebAssembly (WASM) to make these smuggling campaigns even harder to detect.

We analyze via diffing an out-of-bound vulnerability in iOS IOGPUFamily, a kernel driver responsible for handling communication with the GPU (CVE-2025-24257). We will also discuss how this bug is unexploitable on modern Apple kernels, due to heap mitigations.

In digital forensics, AmCache and Shimcache are critical artifacts for understanding system activity and application usage. They provide investigators with metadata about executed programs, installed applications, and file presence, even when files have been deleted. Analyzing these artifacts can reveal malicious activity and and assist in reconstructing accurate system timelines.

We show that Node.js can be weaponized as a post-exploitation loader for your arbitrary code. By swaping out compiled Node.js addons with your own code you can force a legitimate Electron application load and execute your code.

We observed an innovative campaign using the ClickFix attack tactic for cache smuggling. It differs from previous ClickFix variants in that the malicious script does not download any files or communicate with the internet. This is achieved by using the browser's cache to pre-emptively store arbitrary data onto the user’s machine.

In this memory forensics exercise, we will use psscan and psxview to search for hidden processes, dump suspicious processes and uncover macro-enabled documents and rogue executables, and reconstruct a likely attack chain involving phishing, macros, persistence, and possible C2 activity.

This post will guide you on setting up a zero-cost cybersecurity homelab to emulate attacker tactics, techniques, and procedures (TTPs) and hunt them down like a SOC analyst.

We identified a security vulnerability affecting the new Rust-based kernel component of the Graphics Device Interface (commonly known as GDI) in Windows. We detail the methodology of our fuzzing campaign, which targeted the Windows graphics component via metafiles and led to the discovery of this security vulnerability.

In a world with vibe hacking, even if your stack works fine and controls are good; the entire kill chain will be executed before the your first alert fires. We discuss a practical approach to identifying these compressed sequences using tools you already have.

For this part, we will focus on getting a memory dump from hypervisors and cloud platforms. There are two ways to acquire memory from hypervisors and cloud platforms: login into the VM and use the techniques specific to the host OS, or use the built-in tools of the hypervisor or cloud platform to extract memory without interacting directly with the VM.

In this write-up we'll walkthrough the Master File Table(MFT) file records in NTFS (New Technology File System) and see how it exposes the metadata of various files in the volume.

We review how to assess Next.js apps for SSRF, XSS, CSTI, SSTI, CSRF, cache issues, and data leaks. Practical tips, checks, and tools for bug bounty and pentesting.

In this chapter, we're tearing open a Windows memory dump with Volatility 3 to hunt down user activity. As attackers leave breadcrumbs when they mess with a system, our job is to catch them in the act - who logged in, what they executed, and which files they touched.

In this part we will be focusing on how network activity and post-exploitation behavior are captured in RAM. We'll examine artifacts from a live attack involving advance payloads like Meterpreter, suspicious child processes, and unusual outbound connections.

One technique for command execution has been the use of DCOM objects. However, these objects are no longer as effective as they once were. In this short blog post, I'll explain which DCOM objects are still useful across different versions of Windows, which ones no longer work, and how we can fix or adapt them. In the next section, I’ll also share a new DCOM object I discovered some time ago that can still be used for command execution.

We're diving into CVE-2025-9242, an Out-of-bounds Write vulnerability in the WatchGuard Fireware OS.

We discovered a vulnerability in the online poker website ClubWPT Gold which would have allowed an attacker to fully access the core back office application that is used for all administrative site functionality. This vulnerability could have been used to retrieve drivers licenses, passport numbers, IP addresses, transactions, game history, and more.

We have discovered UEFI shells (command-line environments that run before the operating system loads), authorized via Secure Boot and containing capabilities to bypass it on Framework laptops and desktops.

Service triggers can be a pentester's secret weapon, letting low-priv users quietly fire up powerful services like Remote Registry and EFS. In this article, we review Windows service triggers and how to utilize low/no code solutions to activate those triggers.

Most operate through comparable methods, such as Userland Hooking, Memory Scanning, Static Detection, and Heuristic Detection. These techniques all operate in a similar manner from product to product, leading to software agnostic bypasses. This blog post covers some of such EDR evasion techniques I've been using during pentest engagements.

This article will demonstrate why Nushell isn't just another shell, but a transformative analysis environment that can become the most valuable secret weapon in your cyber security arsenal. We'll explore its core philosophy and walk through detailed, real-world security use cases that showcase its power.

We uncovered a new Astaroth campaign that’s taken infrastructure abuse to a new level. Instead of relying solely on traditional command-and-control (C2) servers that can be taken down, these attackers are leveraging GitHub repositories to host malware configurations.

The arsenal of lateral movement techniques was expanded with new methods that abuse the DCOM functionality of various Windows applications. This article will review the various methods of DCOM lateral movement (including some that are yet undocumented), assess their use cases and forensic artifacts and offer methods to detect and prevent the use of these techniques.

In this second part, we investigate the Shin Web Shell techniques such as the calls to the OS, how strings are obfuscated, and how PHP is used to load PHP.

The Windows Filtering Platform (WFP) is leveraged by many EDRs leverage for network traffic control and endpoint isolation. In this article, we explore how WFP can be manipulated to either block an EDR's connection to its cloud backend or bypass its isolation mechanisms. Both cases can effectively "blind" the EDR or reduce its effectiveness.

In this post I walk through three DOM-XSS findings I discovered while hunting on a bug-bounty program: a cookie-scoped bypass of server cookie overwrites, a Chrome innerHTML quirk, and a JSON injection that can overwrite window.

Web Shell in a Nutshell Web shells are nothing new: they have been around for the best part of the last 15 years . In this first part, we will introduce the notion of web shell, give a quick presentation of PHP and look at the loaders, minimal scripts stored on a webserver that will get and run the web shell.

In this post, we'll take a deep dive into the Anti-Sandbox technique that the LummaC2 v4.0 stealer is using to avoid detonation if no human mouse activity is detected. We will also assess the packer and Control Flow Flattening obfuscation to effectively analyze the malware.

We demonstrate how to remotely inject DLLs into another process using Rust and how to use function pointers.