For engineering teams who verify a claim by running it
Project, LLM Security, Adversarial Testing, Agent Design
Adversarial LLM Defense Framework
Public Adversarial Gate — OWASP LLM Top 10 mapped, 171/177 caught on its own synthetic corpus (n=284)
LLM systems have no deterministic audit trail by default. LLM vendors do not ship explainability or validation. Standard SAST/DAST tooling was not built for prompt injection, jailbreak, or tool-abuse classes — the OWASP LLM Top 10 sits outside its coverage.
The Problem
LLM systems have no deterministic audit trail by default. LLM vendors do not ship explainability or validation. Standard SAST/DAST tooling was not built for prompt injection, jailbreak, or tool-abuse classes — the OWASP LLM Top 10 sits outside its coverage.
Approach
Public gate (clone & run): mapped to the OWASP LLM Top 10, 171/177 adversarial caught, 4/107 benign over-blocked, on its own synthetic test corpus (n=284, written for the repo). The counts come from the repo's own evaluate.py; clone it and run it.
Proof Metrics
Want this on your team?
This work maps to 5 qualifying roles. Remote, open to mid-level and senior IC roles.