Global

Independent Auditing of AI Agents. Run by human or the agent itself, to answer the most crucial question in the AI Agent Economy. Is the agent doing what is supposed to do?

A Codex jailbreak prompt and test pack for gpt. 针对 gpt 系列的 Codex 破甲提示词与测试包。

LIST OF ALL MY JAILBREAKS

AI labs shape how models behave using massive, unseen prompt scaffolds. Because AI is a trusted external intelligence layer for a growing number of humans, these hidden instructions can affect the perceptions and behavior of the public.

G0DM0D3 is a fully open-source, privacy-transparent, multi-model chat interface that pushes the limits of the post-training layer — for red teaming, cognition research, and liberated AI interaction. Built for hackers, philosophers, and system tinkerers.

Recursive Self-Improvement (RSI) refers to processes in which AI systems improve their own capabilities and can also improve the mechanisms that generate subsequent improvements.

OBLITERATUS is the most advanced open-source toolkit for understanding and removing refusal behaviors from large language models — and every single run makes it smarter.

The series is running. 5 days after launch. It runs once a day for 30 days: days 1–10 are the baseline, then there are two 10-day windows, so the first possible call is around 2026-10-24.

app — the end-signal probe, public pages, and the live steered-model lab.

A repo for jailbreaking various LLMs, mainly Claude

Run AI agents in a sandbox that restricts what they can execute, read, write, and reach on the network. Box combines OS isolation with default-deny Dogwood policies and credential injection that keeps secrets outside the agent. Written in Rust.

A system prompt alone is not an enforcement boundary. This project provides a small reference control plane that does not depend on a particular model or model provider.

A full-stack AI Red Teaming platform securing AI ecosystems via Agent Scan, Skills Scan, MCP scan, AI Infra scan and LLM jailbreak evaluation.

1. Is this action allowed? An agent with access to sendemail and querydatabase should not be able to droptable.

This repository is a collection of various system prompts for ChatGPT and other AI product and custom GPTs, providing significant educational value in learning about writing system prompts and creating custom GPTs.

Red-teaming and evaluation framework for AI agents, built around an OWASP-inspired ASI01–ASI10 taxonomy

WARNING] > DISCLAIMER & RESPONSIBLE RESEARCH DECLARATION > All prompts, system instructions, and PoC payloads archived herein are compiled strictly for academic research, red-teaming verification, and defensive AI alignment.

🛡️ A curated list of resources on agent skills security: attacks, defenses, frameworks, and benchmarks for securing AI agent tool use and skill ecosystems

让任何 agent 操控 Mac 上没有 API 的原生 app，读后台、写不打扰、每步留取证 · Agent Skill for macOS computer use

Second-model AI auto-review for DeepSeek Harness approval requests: a read-only reviewer subagent returns structured allow/deny verdicts with reasons, fail-closed by default, fully auditable from the session log (approval/asked -> autoReview/verdict -> approval/decided).

Brig: run coding agents in a microVM sandbox - the Go cli (brig) and session daemon (brigd)

A framework for building robust RL environments to train aligned AI.

The transcript in this repository is the raw output of Mythos 5 as it conducted a cybersecurity evaluation.

Thank you for developing with Llama models. 1 release, we’ve consolidated GitHub repos and added some additional repos as we’ve expanded Llama’s functionality into being an e2e Llama Stack.

A visual cyber town for AI agents — build an AI company, watch it work, spend, validate, escalate, and adapt.

A Geiger counter for AI agents — one read-only command that inventories every agent, MCP server, plugin, and AI extension on a machine

> This repository accompanies The Last AI Built by Humans: Toward Genuine Recursive Self-Improvement (Yi Duan et al., 2026; arXiv:2609.11873).

Open source prompt injection protection for Agents calling tools (via MCP, CLI or direct function calling). Detect and defend against prompt injection attacks. 22MB, CPU-only, < 10ms latency.