nemoclaw-configure-inference

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

• Third-party content exposure detected (high risk: 0.80). NemoClaw's onboarding explicitly lets you point the agent at arbitrary OpenAI-/Anthropic-compatible endpoints (e.g., "Other OpenAI-compatible endpoint", local Ollama, vLLM, NVIDIA NIM) and even validates them by sending real inference requests, so the agent ingests untrusted third‑party model outputs that can materially influence its actions.