This advisory introduces ShadowLeak, a novel zero-click indirect prompt injection vulnerability that exploits AI assistants connected to corporate email and web tools. By embedding hidden instructions inside benign-looking HTML emails, attackers induce ChatGPT to perform unauthorized web requests that leak private business data directly from OpenAI’s servers—beyond the visibility of organizational monitoring.

The report explains how the vulnerability works, its broader implications for agentic AI ecosystems, and provides key recommendations for sanitization, logging, permission segmentation, and governance to secure emerging AI-driven workflows.

Download now to learn more.