Offensive Research.
Documentation of my findings in the wild, including vulnerability writeups, deep-dives into protocol manipulation, and novel WAF evasion strategies.
Basilisk: An Evolutionary AI Red-Teaming Framework via SPE-NL.
A deep dive into Smart Prompt Evolution for Natural Language (SPE-NL). I document the methodology for systematic security evaluation of LLMs using genetic algorithms and adversarial mutation engines.
Evolving Adversarial Prompts: The Logic Behind SPE-NL.
A step-by-step walkthrough of the mutation operators, fitness functions, and selection strategies that make Basilisk effective against production LLM guardrails.
Ollama Security Research: Remote Denial of Service via Malformed Model Data.
Two independent denial-of-service vulnerabilities in Ollama's model import pipeline — a GGUF string length panic causing immediate server crash, and unbounded vocab_size causing complete resource exhaustion. Originally reported via Huntr coordinated disclosure.
LiteLLM Security Research: Authentication Bypass, SSRF, & Sandbox Escapes.
Technical analysis of multiple security vulnerabilities identified in LiteLLM: a Pass-the-Hash authentication bypass, a full-response SSRF via custom guardrails, and a sandbox escape using Unicode normalization bypass in input filtering.
Responsible Disclosure
All research published by Rot Hackers follows a strict vulnerability disclosure protocol. We give vendors 90 days to remediate critical flaws before releasing PoC code and technical details to the public domain.