youtube-research

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

• Third-party content exposure detected (high risk: 0.95). Outsider-authored free text is ingested when NotebookLM indexes the user-supplied YouTube URLs (notebooklm source add "{youtube_url}"), which pulls transcript/video text authored by third parties from public YouTube at runtime.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The skill adds external YouTube video URLs at runtime (e.g., youtube.com via "yt-dlp --flat-playlist 'ytsearch20:{topic}'" and "notebooklm source add "{youtube_url}""), and that fetched content is ingested into NotebookLM and directly influences the agent's generated responses.