Hacking AI Agents: Lessons Learned Breaking AI Systems ...with Johann Rehberger скачать в хорошем качестве

Hacking AI Agents: Lessons Learned Breaking AI Systems ...with Johann Rehberger

Informed by years exploring how computer systems break, in this stream we'll discuss machine learning and AI security with Johann Rehberger, reflecting on his journey from traditional pen testing and red teaming to the early days of finding and reporting vulnerabilities in LLM-powered applications and AI agents. We'll talk about the rise of indirect prompt injection, data exfiltration attacks, persistent memory exploits, promptware and prompt-powered command and control. We'll also reference and cover several topics from his “Month of AI Bugs” (https://embracethered.com/blog/tags/m...) research and what these findings reveal about the culture forming around modern AI systems and personal agents. The content for this sessions comes from Johann's research, black hat talks, and [un]prompted session in San Francisco in March 2026. Follow Johann on X here: https://x.com/wunderwuzzi23 Johann's Blog: https://embracethered.com/