mislead

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

• Third-party content exposure detected (high risk: 0.75). Mislead’s required workflow is to design and instrument a honeypot/decoy surface, which at runtime necessarily involves ingesting/processing external attacker-supplied interaction data (e.g., HTTP requests, contact events, captured payloads) into the agent’s LLM context for analysis/alerting—an outsider-authored free-text source.