Every Website With AI Is Vulnerable Pen Testing the AI Attack Surface - Hackers On The Rocks Podcast

Ofir Hamam pours an arak with lemon juice, raises a glass from Tel Aviv, and opens with a claim worth unpacking: every website using AI is vulnerable at some level. Ofir is Head of Offensive Security at Terra Security, where his team does two things: pen tests organizations' AI-powered systems, and uses AI agents to conduct those assessments at scale. In this episode, we dig into both. On the target side — the non-deterministic nature of AI agents means traditional testing logic breaks down. You can't just throw a single quote character at the input and wait for an error. Every query might behave differently. And most engineers building these systems don't yet know how to test them, let alone secure them. Ofir walks through a real example: convincing a medical chatbot's agent that it had switched from production to staging — a completely reasonable instruction to the model — and using that to extract credentials it was never supposed to hand over. On the tooling side — Terra treats AI agents like junior pen testers. Broad knowledge, needs guidance. The key insight: don't give a single agent too much context. More context means more hallucination. Break the process into focused pieces, keep humans in the loop at the critical decision points, and accept that some things still need to be handled deterministically. The closing argument is worth sitting with: stop relying on the model to correct malicious intent. 99% compliance is not a security posture. 🥃 Drink of the Episode: Arak with Lemon Juice (L'chaim 🥂) Links: Ofir Hamam on LinkedIn:   / ofir-hamam-b88358a0   Terra Security: https://www.terra.security/ Website - https://www.desiredeffect.io/hackers-... Spotify - https://open.spotify.com/show/4A11X2N... About The Show: Hackers On The Rocks mixes refreshing conversation with potent cybersecurity insights. Each episode tackles a specific software vulnerability class and a cocktail. Both go down easy. Cheers! The cybersecurity world is a vibrant mix of technical talent and strategic minds. Here at Hackers on the Rocks, we raise a glass to the unsung heroes: the practitioners. While the broader community encompasses diverse roles including governance, risk, and compliance managers, network analysts, and incident responders, our podcast celebrates the hackers on the front lines. It is researchers who are deciphering code, digging into products, and finding unintended consequences within design implementations. Join us as we raise a glass to their expertise and translate their knowledge into clear, actionable insights for everyone. Hosted by Evan Dornbush, founder of Desired Effect — a cybersecurity consultancy helping organizations understand and act on complex threats. 🌐 https://desiredeffect.io