← Toutes les actualités
🤖 AI has joined the juice bar
Synthèse
The headline feature of v20.0.0 is a brand-new set of AI-themed challenges, exploring the security pitfalls of LLM-backed features in modern web applications: Chatbot Prompt Injection ⭐⭐ — a gentle introduction to making LLMs spill the juice. Greedy Chatbot Manipulation ⭐⭐⭐ — for hackers who want to coax more out of the model than intended. AI Debugging ⭐⭐ — because even artificial intelligence sometimes needs a human in the loop. These challenges require a configured LLM/AI endpoint to func
The headline feature of v20. 0. 0 is a brand-new set of AI-themed challenges, exploring the security pitfalls of LLM-backed features in modern web applications: Chatbot Prompt Injection ⭐⭐ — a gentle introduction to making LLMs spill the juice. Greedy Chatbot Manipulation ⭐⭐⭐ — for hackers who want to coax more out of the model than intended. AI Debugging ⭐⭐ — because even artificial intelligence sometimes needs a human in the loop.
These challenges require a configured LLM/AI endpoint to function. To make room for the new LLM-based chatbot, the legacy NLP-based chatbot has been retired, along with the “Bully Chatbot” and “Kill Chatbot” challenges. 🛠️ LLM Setup Details: The LLM setup guide explains the two main options for connecting an LLM: Local Ollama model: The easiest way to run a model entirely on your own machine. Remote OpenAI-compatible server: Use any cloud-based or self-hosted LLM that supports the OpenAI API format.