Vulnerability research · AppSec · Offensive security · AI and MCP
Hi! I'm Fernando, a security researcher based in Brasília, Brazil.
I like understanding where trust between users, applications, and AI systems can fail. These days, I split my time between web application research, offensive security for LLMs and AI agents, and the exploration of Model Context Protocol (MCP) implementations.
My approach is simple: a finding only becomes useful when others can understand and reproduce it. That is why I publish technical analysis, evidence, impact, remediation, and a controlled proof of concept. This work has resulted in ten CVEs so far.
📍 Brasília, Brazil · LinkedIn · English | Português
| # | CVE | Product | Vulnerability | Research |
|---|---|---|---|---|
| 10 | CVE-2026-122158 | InvoiceShelf | Cross-tenant privilege escalation through improper access control | Analysis and PoC |
| 9 | CVE-2026-120343 | CloudClassroom PHP Project | Stored XSS in manageassessment2.php |
Analysis and PoC |
| 8 | CVE-2026-120218 | TypeMill CMS | Stored XSS through an SVG XSLT processing instruction | Analysis and PoC |
| 7 | CVE-2026-116004 | Onyx | Same-origin iframe sandbox bypass and stored XSS | Analysis and PoC |
| 6 | CVE-2026-116000 | PrivateGPT | SSRF in URI-based artifact loading | Analysis and PoC |
| 5 | CVE-2026-110025 | phpwcms | Unrestricted upload with potential remote code execution | Analysis and PoC |
| 4 | CVE-2026-108924 | Kirby CMS | Authorization bypass in dynamic field options | Analysis and PoC |
| 3 | CVE-2026-105950 | Formwork CMS | Stored XSS through a formaction sanitizer bypass |
Analysis and PoC |
| 2 | CVE-2026-90932 | LaraDashboard | Path traversal with arbitrary file deletion/write and potential RCE | Analysis and PoC |
| 1 | CVE-2026-90931 | LaraDashboard | Stored XSS through SVG upload | Analysis and PoC |
Eight of these CVEs were assigned by VulDB and two by VulnCheck.
- MCPEnum — a Go tool that discovers MCP endpoints and enumerates their advertised tools, resources, templates, and prompts without invoking them.
- Pentest-Tools — tools and references I use during authorized penetration tests and labs.
- RedSonnet-5 — agents and commands that support my CTF, HTB, OffSec PG, and DESEC workflows.
- Attack surfaces and trust boundaries in LLMs and AI agents.
- Discovery, enumeration, and abuse of functionality exposed by MCP servers.
- Web application and CMS vulnerabilities, especially access control, XSS, SSRF, path traversal, and file uploads.
- Automation that makes offensive research faster and reproducible.
If you would like to discuss vulnerability research, AI security, or a possible collaboration, reach out on LinkedIn.
Every test and proof of concept published here was developed for systems I own or environments where I had authorization.