September 2026
-
MCP Server Prompt Injection: Proving the Blast Radius
MCP server prompt injection is easy to demo and hard to prove. See how to test to consequence, scope tokens and write findings that hold up.
-
How to Red Team an LLM: A 5-Phase Field Guide
How to red team an LLM in five phases, from mapping tool permissions to proving consequence and writing findings that survive retest. See the method.
-
Automated LLM Red Teaming: What It Misses
Automated LLM red teaming finds jailbreaks at scale but misses tool abuse and excessive agency. See where to automate and what still needs a human.
-
MCP Server Penetration Testing: A Tester’s Checklist
MCP server penetration testing, step by step: map the tool surface, test tool poisoning and tool abuse, prove real impact. See the full checklist.
-
AI Penetration Testing: How to Scope It Properly
AI penetration testing covers two different jobs. See how to scope each one, what to ask a provider, and how to judge findings you can act on.
-
AI-Assisted Pentest Tools: What Works and What Is Hype
The tools that work let the model organise findings that real measurement produced. Ask any tool where its facts come from.
-
Securing AI Agents and Their Tool Use
Agents turn a model’s weaknesses into actions. Secure them by assuming the model will be fooled and making that survivable.
-
How to Red-Team a Large Language Model
Map the permissions and inputs first, then work through instruction bypass, indirect injection, data disclosure, tool abuse and excessive agency.
-
Using Large Language Models for Reconnaissance
LLMs make recon faster by reading and synthesising public information, but they invent facts as readily as they report them. Keep the model out of the fact-finding.