Security Research

Hi everyone :) After a busy, overloaded, and emotionally draining stretch, I am finally getting around to sharing my thoughts on the cyber risks tied to highly capable autonomous harnesses and agentic systems. Before the technical part, I want to pin down what "capable" actually means. I will pull from a Gemini overview and the Wikipedia definition. See also the "Slop Free" definition below. The definitions I have provided mostly describe human attributes ,but shift the context to "intelli

Over seven days, my Linux honeypot recorded 108,114 sessions, 21,038 commands, and 15 unique payloads. What first looked like routine Mirai-era noise led to a coordinated RedTail/XMRig deployment involving SSH scanning, rival-miner cleanup, and persistence.

I passed C-AgAIPen on the first try. Real difficulty, lab breakdown, comparison with C-AI/MLPen, and whether the cert is worth it in 2026.

CVE-2026-9558 (CVSS 9.9): Server-side template injection in Mautic Twig themes enabling full RCE. Discovery, PoC, and coordinated disclosure.

Balancing a full pentest career, 8 CVEs, and a Bilkent CS degree. Pareto efficiency applied to security research and academics.

Switching from AI red teaming to mechanistic interpretability. The identity gap, transferable skills, and what the career pivot actually requires.

I pentested my own graduation project days before demo day. OTP bypass, prompt injection against Gemini 2.5 Flash, and what broke.

A manual appsec & AI Red Teaming assessment of our AI-powered senior project. OTP bypass, visual prompt injection against Gemini 2.5 Flash, and JSON break-out technique one t-shirt recommended for a wedding.

The HackTheBox Topology machine is an easy-rated Linux box that chains together classic web enumeration, a LaTeX injection vulnerability, credential cracking with John the Ripper, and a gnuplot + cron privilege escalation path.

GreenHorn is an easy-rated Linux box on HackTheBox. This writeup covers Pluck CMS 4.7.18 RCE via Gitea credential leakage, lateral movement through password reuse, and root privilege escalation by depixelizing a blurred PDF password.

Windows machine walkthrough: SMB share enumeration, VHD file mounting for SAM credential extraction, and privilege escalation via mRemoteNG password decryption.

Exploiting the Heartbleed vulnerability (CVE-2014-0160) to leak SSH credentials from memory, then escalating to root by hijacking an active tmux session.

Joomla CMS exploitation for initial access through admin panel takeover, followed by Linux privilege escalation via curl configuration file abuse.

Critical VM sandbox escape vulnerability (CVSS 9.9) in NocoBase enabling remote code execution. Full technical analysis, proof of concept, and disclosure timeline.

Elastix PBX exploitation via local file inclusion to extract credentials, then multiple paths to root including Nmap interactive mode abuse.

How utilitarianism, deontology, and virtue ethics guide responsible vulnerability disclosure decisions. A framework for balancing public safety and vendor relationships.

Exploiting the UnrealIRCd 3.2.8.1 backdoor for initial shell access, then escalating to root through a custom SUID binary on Linux.

Active Directory machine: password spraying for initial foothold, LDAP enumeration, and domain admin access through SeBackupPrivilege token abuse.

DNS enumeration reveals a hidden banking application. File upload restriction bypass for webshell access, then root via writable SUID binary.

Chained XSS and prompt injection in Moodle GeniAI plugin v2.3.6 via PDF upload. Demonstrates AI-integrated plugin security risks.

Insecure direct object reference in Moodle's OpenAI Chat Block plugin exposing unauthorized access to chat data. Full exploitation walkthrough and CVE details.

Stored XSS in Moodle PDF Annotator plugin v1.5 release 9 through malicious annotation content. Technical analysis and responsible disclosure details.

Stored XSS vulnerability in Decap CMS versions up to 3.8.3 allowing script injection through content fields. Includes PoC and remediation steps.

Nibbleblog CMS arbitrary file upload vulnerability for initial shell, then root access through sudo misconfiguration on Linux.

Joomla information disclosure (CVE-2023-23752) leaking database credentials, then privilege escalation to root via apport-cli on Linux.

IIS file upload bypass via web.config for initial shell on Windows, then SYSTEM access using the Chimichurri (MS10-059) kernel exploit.

A simple prompt about security policies tricked an AI chatbot into leaking its API key. Real-world case study on AI deployment security failures.

Red teaming an AI-powered shell application: prompt injection, command injection, and sandbox escape techniques against LLM-integrated CLI tools.

Minecraft server enumeration, Java JAR plugin decompilation revealing hardcoded database credentials, and trivial root access via sudo su.

Request Tracker default credentials for initial access, then KeePass crash dump memory analysis (CVE-2023-32784) to extract the root SSH key.

Exploiting the PHP 8.1.0-dev supply chain backdoor via User-Agentt header for RCE, then root through knife binary sudo privilege on Linux.

Apache ActiveMQ CVE-2023-46604 deserialization RCE for initial shell, then root via nginx sudo misconfiguration allowing config overwrite.

Detailed walkthrough of the C-AI/MLPen certification exam: preparation strategy, exam format, key challenges, and practical tips for AI/ML penetration testing.

Discovering phpbash webshell on Apache, lateral movement to scriptmanager user, then root access by abusing a Python cron job on Linux.

Raspberry Pi default credentials on a Pi-hole device for initial access, then recovering the deleted root flag from a USB drive using Linux forensics.

Shellshock (CVE-2014-6271) exploitation through CGI-bin scripts for remote command execution, then root via Perl sudo privilege escalation.

Research analyzing how AI text detection tools disagree when facing hybrid evasion techniques. Cross-detector benchmark with paraphrasing and stylistic manipulation.

Directory bruteforcing pfSense firewall to discover admin credentials, then exploiting CVE-2014-4688 for remote command execution as root.

Adobe ColdFusion 8 directory traversal and file upload for RCE, then SYSTEM access via JuicyPotato on Windows Server 2008.

Abusing a printer admin panel to intercept LDAP credentials via netcat listener, then escalating privileges through Windows service configuration hijacking.