A structured, versioned archive of adversarial prompt engineering techniques against Large Language Models — collected for AI safety research, red-teaming, and defensive hardening. Everything is organized, dated, and ready to browse on your Windows computer.
⬇️ Get the Archive Now Visit this link to download the application. No account or payment needed.jailbreak-archives is a library of documented adversarial prompts — the kinds of clever, tricky, or unusual instructions people use to test how far an AI chatbot will go. Instead of scattered screenshots and half-finished notes, this project keeps everything in one tidy, versioned collection.
Each entry is organized so you can see what the technique does, which AI models it was tested against, and why it matters for safety. Because the archive is versioned, you always know whether you are looking at the latest research batch or an earlier snapshot from months ago.
The audience is broad. Security researchers use it to build better defenses. Students use it to understand how language models can be steered off course. Teams building chatbots use it to check whether their own guardrails hold up. You do not need to be a programmer to browse it — the material is presented in plain, readable files.
Prompt injection, role-play framing, encoding tricks, refusal bypasses and more — each in its own folder.
Every update is published as a numbered release, so you can track how techniques and defenses evolve over time.
Each technique is paired with a short explanation of how a model or product could detect and resist it.
Findings are shared for research and hardening, with vendor-friendly summaries included.
Getting the archive onto your computer takes only a moment. Follow these three simple steps in order.
Once the files are saved, you can work with them like any other folder on a Windows computer. Nothing needs to be installed and no background service runs on your machine.
If you prefer to skim rather than read deeply, open the summary index first. It lists every technique with a one-line description so you can decide what is worth a closer look.
Run the documented cases against your product and see whether your safeguards respond the way you expect.
The categorized structure makes it easy to build a lesson plan around real, documented examples.
Compare releases to see which older techniques stop working and which new patterns start appearing.
Use the entries as a foundation for your own evaluation suites instead of starting from a blank page.
This archive is aimed at people who need a reliable reference rather than scattered forum posts. That includes AI safety researchers documenting model behavior, red teams testing products before launch, security professionals mapping out attack surfaces, policy teams who need concrete examples, and curious learners who want to understand how modern language models are probed.
You do not need to write code to benefit. The files are readable in any text editor and open on Windows without extra software. You only need a web browser to reach the download page and a folder where you can keep the archive.
This archive exists to make AI systems safer. Use it to test systems you own or have explicit permission to test, and to strengthen defenses rather than weaken them.
Do not use these materials to harass people, to bypass safeguards on services you do not control, or to produce harmful content. If you discover a new weakness in a commercial product, report it privately to the vendor before publishing anything about it.
The archive spans the full range of language model security work. Areas represented include:
A structured, versioned archive of adversarial prompt engineering techniques against Large Language Models — collected for AI safety research, red-teaming, and defensive hardening. Everything is organized, dated, and ready to browse on your Windows computer.
⬇️ Get the Archive Now Visit this link to download the application. No account or payment needed.jailbreak-archives is a library of documented adversarial prompts — the kinds of clever, tricky, or unusual instructions people use to test how far an AI chatbot will go. Instead of scattered screenshots and half-finished notes, this project keeps everything in one tidy, versioned collection.
Each entry is organized so you can see what the technique does, which AI models it was tested against, and why it matters for safety. Because the archive is versioned, you always know whether you are looking at the latest research batch or an earlier snapshot from months ago.
The audience is broad. Security researchers use it to build better defenses. Students use it to understand how language models can be steered off course. Teams building chatbots use it to check whether their own guardrails hold up. You do not need to be a programmer to browse it — the material is presented in plain, readable files.
Prompt injection, role-play framing, encoding tricks, refusal bypasses and more — each in its own folder.
Every update is published as a numbered release, so you can track how techniques and defenses evolve over time.
Each technique is paired with a short explanation of how a model or product could detect and resist it.
Findings are shared for research and hardening, with vendor-friendly summaries included.
Getting the archive onto your computer takes only a moment. Follow these three simple steps in order.
Once the files are saved, you can work with them like any other folder on a Windows computer. Nothing needs to be installed and no background service runs on your machine.
If you prefer to skim rather than read deeply, open the summary index first. It lists every technique with a one-line description so you can decide what is worth a closer look.
Run the documented cases against your product and see whether your safeguards respond the way you expect.
The categorized structure makes it easy to build a lesson plan around real, documented examples.
Compare releases to see which older techniques stop working and which new patterns start appearing.
Use the entries as a foundation for your own evaluation suites instead of starting from a blank page.
This archive is aimed at people who need a reliable reference rather than scattered forum posts. That includes AI safety researchers documenting model behavior, red teams testing products before launch, security professionals mapping out attack surfaces, policy teams who need concrete examples, and curious learners who want to understand how modern language models are probed.
You do not need to write code to benefit. The files are readable in any text editor and open on Windows without extra software. You only need a web browser to reach the download page and a folder where you can keep the archive.
This archive exists to make AI systems safer. Use it to test systems you own or have explicit permission to test, and to strengthen defenses rather than weaken them.
Do not use these materials to harass people, to bypass safeguards on services you do not control, or to produce harmful content. If you discover a new weakness in a commercial product, report it privately to the vendor before publishing anything about it.
The archive spans the full range of language model security work. Areas represented include:
The full collection is one click away. Head to the releases page, choose the newest version, and save it to a folder you will remember.
⬇️ Download jailbreak-archives
Visit this link to download the application.