databricks-model-serving

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

• Third-party content exposure detected (medium risk: 0.30). Source text provided by an outsider at runtime is ingested into the required workflow as request.input[*].content via predict()/predict_stream() (e.g., from query_serving_endpoint messages or REST /invocations) and then passed to the agent LLM (self.llm.invoke(messages) or llm_with_tools.invoke(messages)).