Google Workspace’s continuous approach to mitigating indirect prompt injections

Google describes indirect prompt injection (IPI) as an evolving attack that manipulates complex AI applications (e.g., Workspace with Gemini) by embedding malicious instructions into the external data or tools an LLM consumes — potentially influencing model behavior without any direct user input. Google warns this is not a one-off fix: rising agentic automation and diverse content sources create a dynamic threat surface, so they apply a continuous, multi-layered program of defenses (improving model resistance, launching guarded capabilities, monitoring and hardening the application stack) to d