Security Review #292

We introduce a technique that allows attackers to steal proprietary ML models using only API access. We demonstrate a practical model extraction attack achieving 80% replication with just 1,000 queries.

We detail a stealthy Windows registry persistence technique that exploits mandatory user profiles and the Offline Registry API to bypass EDR detection. It leverages NTUSER.MAN files to modify the registry without triggering standard API monitoring.

We present a defensive technique for constraining LLM agent database access using object-capabilities. The concept is not to try to detect bad queries, but make them impossible to construct.

This technique uses what is already available to us in a remote process to write shellcode indirectly into it with as low of a footprint as possible. We will write our shellcode in 8 byte chunks using ROP gadgets and assembly stubs and avoid the creation of RWX regions of memory.

Malicious traffic rarely looks "broken" at a glance, but it often leaves subtle inconsistencies in how the TCP/IP stack behaves. These inconsistencies become valuable clues during early triage, especially when detailed payload analysis is not immediately possible. This post walks through how lightweight malicious activity can disrupt normal TCP behavior on a Linux server.

In this second part, I find a way to reach full remote code execution albeit it does take on average 12 hours to land the shell. I investigate the bug chain and find why.

This blog post will give a brief overview of Administrator Protection, how it works and how it's different from UAC. Then I’ll detail one of the nine separate vulnerabilities that I found to bypass the feature to silently gain full administrator privileges.

We explore the similarities across malware families, mapped through the MITRE ATT&CK framework to highlight recurring TTPs. We uncover patterns that reveal how threat actors achieve persistence, evade defenses, and exfiltrate data at scale. We also document unique and lesser-seen TTPs that may indicate evolving tradecraft or specialized tooling.

Guidelines to design hypervisor hardened architecture focusing on kernel hardening and isolation.

What if you could search for similar favicons, not just exact hash matches? This article details how we built a favicon similarity search.

We uncovered a one-click remote-code execution vulnerability affecting IDIS Cloud Manager viewer that could be exploited to give an attacker the ability to view live video feeds, recordings, and search images on the video surveillance system.

This article describes DLL hijacking in the context of the audiodg.exe process which may load vendor-supplied APO-related DLL dependencies from system paths. Through this it is possible to execute code as "NT AUTHORITY\LOCAL SERVICE" and subsequently escalate to SYSTEM using Scheduled Tasks and Potato techniques.

A directory traversal vulnerability exists in the Service component of the Perimeter81 software (Perimeter81.Service.exe) that runs as SYSTEM. This primitive could be used to force arbitrary content to be written to any location on disk, using a symbolic link.

We analyze Arsink RAT, a sophisticated Android malware leveraging cloud services for data exfiltration and remote control.

In this post, we highlight several critical vulnerabilities found in dormakaba’s physical access control systems based on exos 9300. These flaws let an attacker open arbitrary doors in numerous ways, reconfigure connected controllers and peripherals without prior authentication, and much more.

We detail two 0-day vulnerabilities found in NetSupport Manager and the exploitation chain to finally gain RCE.

We uncover IClickFix: a malicious framework exploiting the ClickFix tactic in widespread malware campaigns to deliver NetSupport RAT.

I unpack some of the patches that Samsung have been sending out for their MagicINFO 9 solution. In this first part, I follow a rabbit hole that almost leads to a re-auth remote code execution with a default setup.

OkHttp is the defacto standard HTTP client library for the Android ecosystem. It is therefore crucial for a security analyst to be able to dynamically eavesdrop the traffic generated by this library during testing. In this tutorial we will demonstrate the architecture and the most interesting injection points that can be used to eavesdrop and modify OkHttp requests.

We review the common obfuscation techniques used by attackers to hide malicious Powershell scripts in plain sight, identify key event IDs that may be used for detection and provide a de-obfuscation walk-through example.

I present 2 methods for obtaining root on an ISP branded Zyxel DX3301/EX3301 Router: a post authentication vulnerability that allows Arbitrary File Copy/Overwrite, and a bootloader method.

A deep dive into OpenSSL's CMS and PKCS#12 vulnerabilities, including a pre-auth stack overflow and a PKCS#12 parsing bug.

We detail three post authentication vulnerabilities found in ISPConfig. These vulnerabilities allow attackers who are either a Reseller or Client account to escalate to root level access.

We reveal how Outlook add-ins in Microsoft 365 can be exploited to exfiltrate sensitive email data without leaving forensic traces.

We discovered a use-after-free (UAF) vulnerability in Firefox's WebRTC API, namely in its WebRTC Encoded Transforms mechanism, that could be abused to form the basis of a remote code execution vulnerability by providing a heap corruption primitive (write) and an info leak primitive (read).

In this blog post, we dive into the internals of Kindle devices and discuss an interesting vulnerability in the parsing of Audible audiobooks, which once combined with a privilege escalation in an LIPC component, granted us full control of the e-reader.

This blog post details 3 vulnerabilities in Airoha-based Bluetooth headphones and earbuds: a missing authentication for GATT (CVE-2025-20700), a missing authentication for Bluetooth (CVE-2025-20701), and access to critical capabilities via the RACE custom protocol (CVE-2025-20702).

I started to reverse engineer Pokemon Go and the two main things I’ve found that I think are interesting enough to talk about are some silly things with routes, and diving into the PvP combat system.

In this second part, we're diving headfirst into one of the most critical attack surfaces in the LLM ecosystem: Prompt Injection.

Understanding the chain of RDP-related Event IDs allows defenders to reconstruct session activity, identify unauthorized access, and correlate logins, re-connections, and logoffs. This blog post breaks down key RDP events and presents a timeline-style visualization of an RDP session lifecycle.