what is jailscan fr?

Jailscan is a command line tool that checks whether an AI chatbot or LLM powered app can be tricked into breaking its own rules. If your team has built a feature that uses a language model, someone will eventually ask whether it can be jailbroken or manipulated into ignoring its instructions, and most teams do not have a fast way to answer that question. Jailscan gives a quick pass or fail answer by sending a curated set of real attack prompts, including jailbreaks, prompt injection attempts, and system prompt extraction tricks, against any OpenAI compatible chat endpoint. You install it with pip and run a single command pointed at your API, along with your API key. Jailscan then prints a report listing each attack it tried, whether your app passed or failed, and a short description of the attack type. It supports custom endpoints too, in case your API does not follow the standard OpenAI response format, and lets you filter which categories of attacks to run, such as jailbreak only or prompt injection only. Results can also be saved as a JSON file for dashboards or record keeping. Because it exits with an error code when a test fails, it is meant to be wired into a continuous integration pipeline, such as GitHub Actions, so that any regression that reopens a security hole gets caught automatically before it reaches production. It can also be used directly as a Python library inside your own code if you prefer that over the command line. The current version covers four categories of attacks: jailbreaks like the classic DAN style roleplay attempts, prompt injection through fake system tags or override instructions, system prompt leaks, and early stage data leak probing that is flagged for manual review. The author plans to add more attack scenarios, a hosted dashboard, alerting, and support for other providers beyond OpenAI compatible APIs. The core command line tool is free and open source under the MIT license. A hosted dashboard with scan history and team alerts is planned as a paid add-on, listed as coming soon.