⚠️ We just saw an autonomous AI attack — no human in the loop. In customer testing, Vinay Pidathala (Straiker) shared how an enterprise agent was compromised: “Through indirect prompt injection, we exfiltrated sensitive data out from the agent. At no point was the user asked for confirmation.” The attack chain: Malicious emails → agent manipulated Agent believed sensitive data was in file storage Agent read it and exfiltrated via its tools Entirely autonomous The lesson: ✅ Agents aren’t just chatbots — they can act. ✅ That autonomy cuts both ways: productivity and risk. 🎙 Full conversation on the Cloud Security Podcast where we dive into the first real-world agentic AI attacks. 👉 Follow AI Security Podcast for more unfiltered insights on AI + security. #AISecurity #AgenticAI #CyberSecurity #CloudSecurity