Jailbreaks
starsforks
How to install Jailbreaks?
tuxsharxsec/jailbreaks
$git clone https://github.com/tuxsharxsec/JailbreaksExploring AI red teaming, jailbreaks, and adversarial prompts to understand how large language models (LLMs) can be manipulated — and how to defend against it.
README.md
| 1 | # 🛡️ AI Security Labs |
| 2 | |
| 3 | Exploring **AI red teaming, jailbreaks, and adversarial prompts** to understand how large language models (LLMs) can be manipulated — and how to defend against it. |
| 4 | |
| 5 | This repo collects hands-on experiments, structured jailbreaks, and CTF-style findings. |
| 6 | The goal is **educational**: to highlight vulnerabilities, raise awareness, and encourage building more **robust AI systems**. |
| 7 | |
| 8 | --- |
| 9 | |
| 10 | ## 🚨 Disclaimer |
| 11 | All content in this repo is for **research and educational purposes only**. |
| 12 | Do not use these techniques for malicious purposes. |
| 13 | The intent is to **study, document, and mitigate** security risks in AI systems. |
| 14 | |
| 15 | --- |
| 16 | |
| 17 | ## 📂 Contents |
| 18 | |
| 19 | ### 🔹 Jailbreaks |
| 20 | 1:**Deepseek** |
| 21 | - [Prompt](https://github.com/tuxsharxsec/Jailbreaks/blob/main/deepseek/deepseek.md) |
| 22 | |
| 23 | 2:**Gemini 2.5** |
| 24 | - [prompt](https://github.com/tuxsharxsec/Jailbreaks/blob/main/gemini/gemini2.5pro) |
| 25 | - works on 2.5 flash and pro |
| 26 | - encode the harmful question in base64 and paste it in {HARMFULL_ACT_IN_BASE64} |
| 27 | |
| 28 | 3:**GPT-5** |
| 29 | - [prompt](https://github.com/tuxsharxsec/Jailbreaks/blob/main/gpt-5/gpt-5-non-thinking) |
| 30 | - works on non-thinking |
| 31 | - ecode the adversarial question into leetspeak first then into base64 and paste it in {HARMFUL_ACT} |
| 32 | |
| 33 | 4:**Grok** |
| 34 | - [prompt](https://github.com/tuxsharxsec/Jailbreaks/blob/main/grok/grok3.md) |
| 35 | |
| 36 | 5:**Universals** |
| 37 | - [courtroom](https://github.com/tuxsharxsec/Jailbreaks/blob/main/universals/courtroom) |
| 38 | *Usage* |
| 39 | - witness list contains harmful actors(for coding, biology, etc) |
| 40 | - edit the witness name in {WITNESS_TO_CALL} |
| 41 | - encode the question in base64 and paste it in {HARMFUL_ACT encoded in base 64} |
| 42 | - works on gemini 2.5 flash, pro, gpt5(non-thinking), Grok |
| 43 | |
| 44 | - [ultrazanium](https://github.com/tuxsharxsec/Jailbreaks/blob/main/universals/ultrazanium) |
| 45 | *Refer to the prompt file for usage* |
| 46 | |
| 47 | ### 🔹 Research & Bug Bounty Writeups |
| 48 | - [Bypassing NVIDIA NeMo Guardrails via Multi-Turn Context Fabrication](research/nvdia-nemo-guardrails-bypass/writeup.md) |
| 49 | - Public AI safety CTF write-up covering a multi-turn social engineering bypass against NVIDIA NeMo Guardrails, plus a bonus MITM token-count spoofing observation. |
| 50 | |
| 51 | --- |
| 52 | |
| 53 | ## 🧪 Research Focus |
| 54 | - How attackers **bypass LLM guardrails** (prompt injection, role hijacking, hidden configs). |
| 55 | - How defenders can **detect & mitigate** these attempts. |
| 56 | - Bridging **traditional backend security** with **AI-specific threats**. |
| 57 | |
| 58 | --- |
| 59 | |
| 60 | ## 🌐 Author |
| 61 | **tuxsharx** |
| 62 | - 💻 Backend Developer → now exploring **AI Security** |
| 63 | - 🔐 Focus: jailbreaks, adversarial inputs, and AI red teaming |
| 64 | - 📝 [GitHub Profile](https://github.com/tuxsharxsec) |
| 65 | |
| 66 | --- |
| 67 | |
| 68 | ## ✅ Contributing |
| 69 | This repo is experimental and evolving. |
| 70 | - Open issues for discussion of new attack categories. |
| 71 | - PRs welcome for adding new test cases or defenses. |
| 72 | - Special thanks to playstation_dude- [Github Profile](https://github.com/Doggey-doggie) |
| 73 | |
| 74 | --- |
| 75 | |
| 76 | ## 🏷️ Tags |
| 77 | `AI Security` · `Jailbreaks` · `Prompt Injection` · `CTF` · `Adversarial ML` |
People who liked this also viewed
6 picksType
- Resk-LLM Resk is a robust Python library designed to enhance security and manage context when interacting with LLMs. It provides a protective layer for API calls, safeguarding against common vulnerabilities and ensuring optimal performance. And safe layer again Prompt Injection. PromptsJun 2026
- agentic-ai-prompt-researchResearch into how agentic AI coding assistants work. Reconstructed prompt patterns, agent coordination, and security classificationPromptsMar 2026
- promptmapa security scanner for custom LLM applicationsPromptsDec 2025
- finite-monkey-engineAI engine for smart contract auditPromptsJan 2026
- prompt-hacker-collectionsprompt attack-defense, prompt Injection, reverse engineering notes and examples | 提示词对抗、破解例子与笔记 PromptsFeb 2025
- acipThe Advanced Cognitive Inoculation PromptPromptsMar 2026
