AI-Native Security Is a Must to Counter AI-Based Attacks

**Executive summary:** Panelists at Nvidia's GTC warned that agentic AI frameworks like OpenClaw can create a new persistent attack surface—spawning subagents that scan file systems, access personal data, remain dormant for long periods, and activate later—and recommended AI-native defenses (dynamic access control, identity separation for agents, audit trails, knowledge/context graphs, and secure SDLC) to mitigate these risks.