AI-103 Practice Question 87: Video Content Moderation
Microsoft Azure AI Apps and Agents Developer Associate (AI-103) · Test 9 · question 7 of 10
A user-generated video platform needs to automatically flag uploaded clips that contain adult or racy visuals, and separately flag clips whose spoken transcript contains explicit language, so moderators can jump straight to the flagged timestamp instead of watching the whole file. The team has no labeled training data and doesn't want to build or maintain a custom classifier.
Which capability fits, with the least custom work?
Show answer and explanation
Answer:
Video Indexer ships two purpose-built moderation models as documented insights: visual content moderation "detects adult and racy visuals," and textual content moderation "detects explicit text in the audio transcript." Both are listed as standard insight types, which means each flagged occurrence comes back as a timestamped instance in the video (exactly what lets a moderator jump to the moment in question) with no model to train or label.
- A is wrong because a video analyzer's custom fields are answered by a generative model describing whatever you ask it to describe; that isn't the same as a purpose-built, responsible-AI-reviewed moderation classifier, and Microsoft doesn't document it as a safety capability. - C is wrong because Content Safety's image endpoint analyzes a single still image per call. Feeding it a video file directly isn't supported; the team would have to write its own frame-extraction and timestamping code first, which is exactly the custom work they're trying to avoid, and it does nothing for the audio transcript. - D is wrong because Custom Vision requires collecting and labeling a training set of accepted and rejected footage and maintaining that model over time; the "no labeled data, no custom classifier" constraint rules it out directly.
Reference: What is Azure AI Video Indexer?