Google Workspace’s continuous approach to mitigating indirect prompt injections

Google warns of indirect prompt injection (IPI) as an evolving threat to complex AI applications (e.g., Workspace with Gemini) where adversaries inject malicious instructions into the data or tools an LLM uses—potentially without any direct user input. IPI is framed as a persistent, dynamic risk amplified by agentic automation and diverse content sources; Google describes a continuous, defense-in-depth program to reduce risk through improved model resistance, content and tool sanitization, monitoring/detection, adversarial testing, and staged/safer feature rollouts.