watch-video

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

• Third-party content exposure detected (high risk: 0.75). Yes—when the user supplies an outsider video URL (e.g., YouTube/Loom/Vimeo/Riverside/X/IG/TikTok), the workflow fetches platform-provided transcripts and/or downloads the video and then feeds the resulting transcript text (outsider-authored speech captions) into the LLM during the visual/multimodal “Per-frame batch prompt” and summary synthesis.