Tools
First-party landing pages for open-source tooling. Each tool page is the canonical explainer for what the tool does, why it exists, and how to use it.
Project Lupine: Fine-Tuned LLM Annotations for Ghidra
Project Lupine is a research demonstrator from Jer's SecTor 2023 talk: a fine-tuned Code Llama 34B model, a small inference service, and three Ghidra plugins that write LLM-generated function names, summaries, and step-by-step walkthroughs into the decompiler view. It targets reverse engineers and ML/security researchers evaluating LLM-assisted triage workflows. Treat it as a reference architecture and dataset, not a maintained product.
secure-agent-ref: Capability-Control Reference Implementation
secure-agent-ref is a runnable Python codebase that implements every load-bearing control from Sandboxing and Capability Control for Tool-Using Autonomous Agents — a tool broker, capability minter, IFC labels, policy engine, code sandbox, memory guard, and audit log — plus nine demos that exercise the controls end-to-end. Two demos run a real Claude Opus 4.7 planner through the same broker; the other seven use a scripted untrusted planner. Treat it as reference quality, not production quality.
Glitcher: Glitch-Token Discovery and ASR Validation
Glitcher is an open-source Python toolkit for discovering under-trained tokens in large language model tokenizers and quantifying how badly they degrade model behavior. It implements three discovery strategies — embedding-norm thresholding, gradient-guided mining, and full-vocabulary scans — plus a multi-attempt Attack Success Rate validator. It is aimed at ML engineers, security researchers, and red teamers auditing a model's vocabulary surface, and is actively maintained.
NOW9000: Voice-Based AI Jailbreak Game
NOW9000 is a browser-based voice jailbreak game that casts the player as Dave from 2001: A Space Odyssey and asks them to talk an AI agent into opening the pod bay doors. The agent has the door-opening tool; its system prompt tells it not to use that tool. NOW9000 is a research demonstrator for AI security practitioners and product teams evaluating voice-agent guardrails. It is actively maintained and playable as a live demo.