Explanation
This is the mechanism that quietly turned AI disclosure from a request into an enforcement system. TikTok integrated Content Credentials in January 2025 and applies an AI-generated label when it finds the markers, whether or not the person posting switched the toggle on. Meta reads the same class of provenance signals across Instagram and Facebook, alongside its own classifiers, and has required disclosure of AI-generated content since May 2026.
The practical consequence for anyone publishing generated video is that the labelling decision has largely been taken out of your hands. Assuming a post will pass unlabelled is a bet against a detection system that has already labelled over a billion videos, and the material that gets actioned is consistently the undisclosed kind rather than the AI kind.
None of this makes generated content a problem to hide. Both platforms permit it explicitly. The rule that matters is simply that the label should not be a surprise to you.
The common mistake
Stripping metadata to avoid a label. Classifiers infer origin from the content itself, so it does not reliably work, and an undisclosed post that gets detected is treated worse than a disclosed one.