Skip to the writing

agent-security

2 posts.

MCP Attacks on Xata and GitHub MCP: Read-Only Bypass and Issue InjectionDay 8 / 100 · AI cybersecurity
Two MCP lab walkthroughs: a read-only SQL bypass, a GitHub issue injection, and the model refusals and fixture changes that limited what the results established.
MCP security: when prompt injection gets a toolDay 5 / 100 · AI cybersecurity
MCP makes it easier for AI systems to use external tools. It also connects model behaviour to permissions, private data, and real actions.

Ryan Sacatani

Simply curious about the world, constantly building and breaking things for fun.

sacataniryan1@gmail.com ↗

BrowseBrowse topics

Browse

Cloud Security3
- Google Cloud Certifications2
  - Google Cloud Cybersecurity: The Labs Made It Stick1
  - How I Passed Google’s Professional Cloud Security Engineer Exam Using AI1
- MCRTA1
  - I Passed MCRTA - A Beginner Cloud Red Team Exam1
Offensive Security25
- CRTO25
Web Security3
- Labs & Write-ups2
  - Web LLM attacks1
  - Web LLM attacks, deeper into indirect prompt injection1
- Testing Methods1
  - The Evolution of Penetration Testing1
AI Security11
- Models & Applications2
  - An interactive playground for LLM security testing1
  - Poisoning an AI assistant's memory1
- Agents & Tool Access5
- Sandboxes & Control1
  - How to Sandbox an AI Agent: Linux Boundaries and Real Risks1
- Experiments & Builds1
  - How to Build an Agentic Vulnerability Research Harness (Like the Pros)1
- Incidents & Research2
  - Agent Swarms Are Here, and I Don't Think We're Ready1
  - Deeper into the OpenAI HACKING & HACKED!!1
Mobile Security1
- Application Security1
  - Building and breaking mobile applications1
Systems & Architecture2
- Infrastructure & Linux1
  - Why Kubernetes Exists: From One Container to Forty Servers1
- Secure Development1
  - Building a DevSecOps Pipeline with GitHub Actions1
Career & Life2
- Security Career2
  - My first offensive security job1
  - Senior in 18 months, and a red team built from nothing1

RSS ↗ · Brain