Meta's Oversight Board determined on September 17, 2026, that the company's policies on AI-manipulated content and deepfake labeling are inadequate in two specific cases. The independent board ordered the removal of two videos and issued clear demands to strengthen abuse prevention.
The essentials
- Two cases: The Board ordered removal of an AI video targeting a young Muslim woman in Europe and a video falsely portraying a Scottish Labour councillor discussing refugees
- Definition too narrow: Meta's concept of "unwanted manipulated imagery" does not capture instances where private individuals are falsified in their words and actions
- Labeling insufficient: Videos should have been marked with the "High Risk AI" label but were not
- Demands: Expanded labeling rules, higher penalties for repeat sharers, and greater transparency on AI labeling practices
The first video: Targeting a private individual
In the first case, Meta failed to remove an AI-generated video targeting a young Muslim woman in Europe. The Oversight Board ordered its deletion and criticized Meta's understanding of abuse. The board majority found that the platform's definition of "unwanted manipulated imagery" needs reconsideration – it is currently too narrow.
"Meta defines mass harassment campaigns too narrowly, according to the majority, and the company should adopt new measures to reduce the burden on victims reporting them."
The Board noted that mass harassment campaigns are also defined too narrowly. Meta's reporting systems place excessive burden on victims – a structural problem that extends beyond individual content.
The second video: False statements by a politician
The second case involved an AI video falsely portraying a Scottish Labour councillor making a specific statement about refugees. The Board found that this video violated Meta's Hateful Conduct policy and ordered its removal.
However, an additional problem emerged: the video should have been labeled with a "High Risk AI" tag but was not. The Oversight Board sees this as a systemic error in Meta's labeling practices.
What Meta must do now
The board formulated concrete recommendations:
| Measure | Goal |
|---|---|
| Expanded "High Risk AI" labels | Capture more situations requiring AI content labeling |
| Stronger distribution limits | Slow and limit spread of deceptive AI content |
| Higher penalties | Stronger sanctions for accounts repeatedly sharing manipulated content |
| Transparency report | Disclose when and how often AI labels are applied |
The Oversight Board functions as an independent review body and cannot force Meta to implement these recommendations – but its decisions carry significant weight for platform credibility and compliance strategy.
What this means for German companies
The Oversight Board's criticism shows that even major platforms lag in deepfake prevention. For German companies present on Meta platforms or using AI-generated content, this signals: rules will tighten. Those sharing manipulated content – intentionally or negligently – must expect stronger consequences. It also reveals that technical detection and labeling of AI content remains unreliable. This is also relevant for the EU AI Act, which mandates similar transparency and labeling requirements.
Sources
Editorially owned by Ideal Syka. Sources and method: Newsroom & method. Tips and corrections: ai@i6eal.de.
