One video came from India. A teacher was seen scolding and hitting a student in a classroom. The child’s face was mostly covered with a blur. The second video was from France. It showed a teacher hitting a crying child while other children watched. All faces were obscured.
Meta took down both videos under its child abuse policy, which does not allow any content showing child abuse, even when the children are not identifiable. The company later reversed its position on the French case after deciding the video served the public interest. That video was restored with a warning screen and restricted to users over 18.
The Oversight Board reviewed both cases. It agreed with Meta’s decision to keep the French video online. It disagreed with the removal of the Indian video and directed Meta to restore it under the same conditions.
According to the board, Meta’s current policy does not leave room for content that shows abuse without revealing a child’s identity. The board said this blocks important content that highlights wrongdoing in schools. It recommended a narrow exception in the policy, where such videos can be shared if children are not named, their faces are not visible, and the goal is to raise awareness or report abuse.
The board also advised Meta not to penalize users when content is posted for this purpose and removed later, as long as it is clear the intent was not harmful.
Some board members disagreed. They said keeping the videos up could still harm the children, even if they are blurred. They called for removal in both cases.
The recommendation now puts pressure on Meta to allow certain videos of abuse that are shared responsibly and in the public interest.
The cases highlight how Meta’s heavy reliance on automated moderation, with little human oversight, continues to trigger unnecessary censorship, leading to blocked accounts, public backlash, and customer loss, raising the need for better strategies grounded in human judgment.
Notes: This post was edited/created using GenAI tools. Image: DIW-Aigen.
Read next: Google Adds Storybook Feature to Gemini While DeepMind Upgrades Real-Time AI Worlds