.joward

If you know someone who knows someone who knows someone at DEFCON looking to trade workshops, pass this on: I am looking for “From Prompt to Protection: A Practical Guide to Building and Securing Generative AI Applications” workshop ticket, I can trade for “Advanced Ghidra Scripting & Automation.” 😎 Thanks for the help! #defcon33 #defconworkshop #defcon

🚨 It’s happening tomorrow at @_leHACK_ ! I’ll be giving my talk on 𝐀𝐈 𝐑𝐀𝐆 𝐒𝐲𝐬𝐭𝐞𝐦𝐬, where I’ll introduce a technique I developed to extract hidden information from RAG systems, called 𝐔𝐩 𝐚𝐧𝐝 𝐃𝐨𝐰𝐧 𝐓𝐞𝐜𝐡𝐧𝐢𝐪𝐮𝐞. 🧠 In this talk, we’ll dive into the inner workings of Retrieval-Augmented Generation (RAG), expose potential attack vectors, introduce the Up and Down Technique, and walk through practical mitigation strategies. 📍 Join me at @_leHACK_ 🗓️ 27 June 🕐 11:15 AM 🔗 Here is the list of all talks: lehack.org/2025/tracks/confe… #leHACK #AISecurity #cybersecurity

I recently attended a few talks on AI security, and honestly, it felt pretty uncomfortable when some speakers tried to explain key concepts. Many presenters clearly lacked a solid foundation in data science and AI engineering. They didn’t fully understand how models and RAG actually work. Often, they were explaining topics beyond their expertise, mixing concepts up, and spreading misinformation about how things really work. Please, if you’re preparing a talk and decide to discuss topics outside your comfort zone, it’s your responsibility to understand every detail you put on your slides. Otherwise, it’s obvious that you’re just trying to ride the AI hype without genuine understanding.