deep-research

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.85). The required workflow uses Gemini Deep Research to perform web research and then ingests the agent’s returned free-text output (outputs[-1].get("text") / SSE content.delta text) into the program’s LLM-facing context/output, which can include outsider-authored web content and thus enables indirect prompt injection.