Hacking the Claude Ecosystem. Jailbreaking Opus and how to protect it

How Claude-powered systems actually get hacked. We'll go straight into the attacks people are running right now: prompt injection, jailbreaks, agent and tool exploits, and what red-teaming the Claude ecosystem really looks like.

🍕 18:00 – 18:30 · Gathering, Pizza & Drinks
📣 18:30 – 18:45 · Welcome + Latest Claude Updates · Organizers
💻 18:45 – 19:15 · JavaScript Is All You Need: Exploiting Agentic Workflows for API Key Theft · Eden Katz, GenAI Security Tech Lead @ Alice.
Building with AI APIs? This one matters. Alice's research shows that API key creation on platforms like Anthropic relies on cookie-based auth alone — so a malicious browser extension or a compromised agent skill can silently mint a new key and ship it elsewhere. No MFA, no warning, no expiration. As agents get trusted with autonomous code execution and client-side access, the line between safe tool use and serious vulnerability collapses. JavaScript really is all you need.
👾 19:15 – 19:45 · Securing coding agents: Unpacking Threats and Defences · Bar Kaduri, Principal researcher @ Capsule Security
AI coding agents like Claude Code, Copilot, and Cursor now write, execute, and ship code with unprecedented autonomy, often straight to production. But this privileged access opens a unique attack surface, as recent incidents like prompt poisoning and deleted databases reveal. This session breaks down the real-world threats facing coding agents and delivers practical, actionable defences for securing your AI-assisted development pipeline.
🍻 19:45 – 20:00 · Drinks & Geek Out