OS Command Injection in FileRun's thumbnail generation - authenticated (CVSS v4 8.7) or via public 'file request' weblinks (CVSS v4 9.2). The extractors wrap file paths in double quotes and pass them to exec(), but the filename sanitizer allows $(){}` - enough for command substitution. The weblink path uploads a malicious filename via anonymous upload, then self-triggers thumbnail generation…
An unauthenticated PHP object injection in Aimy Captcha-Less Form Guard for Joomla, where a repeating-key XOR published alongside its own ciphertext turns every protected form into an unserialize() sink and, on Joomla 3.9 through 5.2.1, into remote code execution.
An unauthenticated SSRF in Monsta FTP 2.14.4. The product ships a real SSRF blocklist that correctly blocks 127.0.0.1, 169.254.169.254 and RFC1918, then forgets that ::ffff:169.254.169.254 is the exact same host. One missing normalization, both gates bypassed, and a static AAAA record is all it takes.
A critical unauthenticated RCE vulnerability in NVIDIA's GEN3C project. Two FastAPI inference endpoints deserialize raw HTTP POST bodies with pickle.loads() without any authentication, giving instant code execution to anyone with network access.
A missing throw keyword in FOSSBilling's API router exposes the entire admin surface to unauthenticated attackers. Combined with an unsandboxed Twig SSTI that leaks the full DI container, this leads to arbitrary SQL execution, admin takeover, and unauthenticated RCE via malicious extension installation.
A Jinja2 sandbox bypass in NetBox allows low-privilege users to achieve remote code execution via the RenderTemplateMixin environment_params finalize parameter, affecting both ExportTemplate and ConfigTemplate.
Unauthenticated remote code execution in OpenCATS through unsanitized input in the installer AJAX endpoint, allowing PHP code injection into config.php.
A critical unauthenticated RCE vulnerability in ktransformers' balance_serve backend. A ZMQ ROUTER socket binds to all interfaces and proxies messages to worker threads that deserialize them with pickle.loads() - no authentication, no validation.
A critical unauthenticated RCE vulnerability in HuggingFace's LeRobot project (21.5k stars). The gRPC PolicyServer deserializes attacker-controlled data with pickle.loads() in two RPC handlers, allowing instant code execution without authentication.
A local code execution vulnerability in Microsoft's tensorwatch. Calling tw.Watcher() - the first line in every README example - silently creates a ZMQ REP socket on localhost that deserializes incoming messages with pickle.loads(). Any local user on the same machine gets code execution.
Instagram's 'seen' indicator is a separate GraphQL call that any browser extension can block. It's been this way since 2019. Now Meta wants to charge $2/month for it.
68 modules in 2.5 years. Here's what the official docs don't tell you about writing Metasploit modules - from finding targets to surviving code review.
Swap files from Vim and nano can leak usernames, hostnames, and sensitive data in git repos. Even after deletion, the blob stays in git history forever. Here's how to find them and how to actually clean them.
CeWL has been the default wordlist generator for 10 years. CeWL AI crawls HTTP, FTP, SFTP, SMB, and S3 targets, feeds context to an LLM, scans for secrets with 800+ trufflehog detectors, and dumps files - all from one binary.
The loginWithMailLink endpoint in Xboard and V2Board returns the magic login link in the HTTP response body, allowing unauthenticated attackers to take over any account - including admin.
A technique for extracting PostgreSQL data through arbitrary file read vulnerabilities - without credentials, without SQL access, without knowing the schema. Full auto-discovery via system catalogs.
Critical vulnerabilities in Windmill: unauthenticated path traversal leading to RCE, plus an authenticated SQL injection enabling full privilege escalation (operator → super admin → root RCE). Affects standalone Windmill and Nextcloud Flow.
How I reverse-engineered the complete USB HID protocol of the ITE 8910 keyboard controller from a Windows DLL and .NET executable, and contributed per-key RGB support with 14 modes to OpenRGB - the first implementation for this chip on Linux.
A critical unauthenticated RCE vulnerability in OmniGen2's reward server infrastructure. The Flask-based servers deserialize raw HTTP POST bodies with pickle.loads() without any authentication, giving instant code execution to anyone with network access.
A critical unauthenticated RCE vulnerability in sglang's ZMQ transport layer for disaggregated serving. ZMQ PULL sockets bind to all interfaces and deserialize messages with pickle.loads() - no auth, no validation. Distinct from CVE-2025-10164 which only covers the HTTP API.
Five vulnerabilities across SPIP plugins: two SQL injections, two RCE (one unauth, one auth), and reflected XSS. Same template engine, same mistakes, different entry points.
From VulnCheck advisory to working PoC in 30 minutes. Full AI-assisted reversal of CVE-2025-71243, an unauthenticated PHP code injection in SPIP's Saisies plugin affecting versions 5.4.0 through 5.11.0.
How one API designed for disability access became the foundation of a $145M surveillance industry. A proof-of-concept implant demonstrates the full attack chain: silent permission escalation in 2.4 seconds, contextual keylogging, see-through overlays, network toggle, self-hiding persistence, and an embedded Linux terminal with apt - all from a single accessibility toggle, no root required.
CVE-2026-26220: A critical unauthenticated RCE vulnerability in LightLLM's PD disaggregation system. Two WebSocket endpoints deserialize binary frames with pickle.loads() without authentication, and the server explicitly refuses to bind to localhost - it's always network-exposed.
A critical unauthenticated RCE vulnerability in manga-image-translator. Two FastAPI endpoints deserialize raw HTTP POST bodies with pickle.loads(), and the nonce-based authentication is bypassed because the default value is an empty string - which is falsy in Python.
A comprehensive security audit of AVideo revealing 10 vulnerabilities including a critical unauthenticated RCE that chains cryptographic weaknesses, predictable salt bruteforce, and an eval() vulnerability to achieve complete server compromise in under 10 seconds.
A critical vulnerability in Streama allows authenticated users to write arbitrary files through a combination of Server-Side Request Forgery (SSRF) and Path Traversal. This write-up covers the root cause analysis, exploitation flow, and the vendor's comprehensive fix.
Deep dive into bind‐shell deployment, payload experiments, and a new ‘time_conf’ primitive for stealthy, persistent root access without reboot or UI lockup.
Several critical vulnerabilities discovered in Xorcom CompletePBX 5.2.35, including authenticated file disclosure, remote command execution as root, file deletion, and reflected XSS. This write-up details the black-box methodology, PoCs, and patch timeline.
Two missing checks inside the Open Contributions plugin let any fresh WordPress subscriber escalate to contributor and read arbitrary files — including the CTF flag — without brute-forcing a thing.
Analysis of two unauthenticated XSS vulnerabilities and silently patched issues in Vembu BDRSuite up to version 7.6.0. Includes PoCs, API traces, patch timeline and commentary.
WPProbe is a lightweight tool that leverages the WordPress REST API to detect installed plugins passively. This post covers its design, use cases, technical constraints, and how it helps avoid noisy scans.
Two Stored XSS vulnerabilities discovered in MagnusBilling 7.x, including one unauthenticated, initially spotted during a CTF. This write-up covers the discovery process, PoCs, impact, patch timeline, and thoughts on the overall codebase.
An in-depth forensic analysis of how a seemingly legitimate Proof-of-Concept (PoC) for CVE-2020-35489 turned out to be a cleverly disguised malware. This blog post details the attack vector, payload deobfuscation, Indicators of Compromise (IoCs), and the steps taken to analyze and neutralize the threat.