patent

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

• Third-party content exposure detected (high risk: 0.75). High indirect injection risk: Workflow 1/2/3 instructs executing searches on public web patent databases (e.g., Espacenet/Google Patents/WIPO) and then processing the resulting patent text/metadata into JSON/markdown that the agent feeds into the LLM context; those pages are outsider-authored free text at runtime.