Meta told to take down UK deepfakes
Meta ordered to remove UK deepfakes as oversight board criticises ‘inadequate’ safeguards
Sunday, 20 September 2026
Warm-up
- What risks do AI-altered videos create for public debate?
- When should a platform remove content instead of labelling it?
- How do labels or warnings change your behaviour online?
Vocabulary
- oversight
- supervision of an organisation’s actions by an independent authority
- deepfake
- an AI-altered video or image that makes someone appear to say or do things they did not
- safeguard
- a rule or measure designed to protect people from harm or abuse
- synchronised
- matched in timing, such as audio aligning with lip movements
- harassment
- hostile behaviour that targets and intimidates a person or group repeatedly
- transparency
- clear openness about decisions, data or processes so people can understand them
- algorithm
- a set of rules a computer uses to decide what content to show
- demote
- to reduce the visibility or priority of something
Reading
Meta’s oversight board ordered Facebook to remove two UK‑linked deepfakes and said the company’s safeguards were “consistently and fundamentally inadequate” for today’s rapid spread of AI‑manipulated imagery. One clip falsely portrayed a Labour councillor in Scotland making inflammatory remarks about refugees; the audio was not fully synchronised with the councillor’s facial movements, a strong sign of fabrication. Meta left the post up and did not apply any AI label.
A separate ruling targeted an AI video that mocked a young Muslim campaign volunteer in Europe who promotes menstrual health education. Versions of the clip drew tens of millions of views, and the board said it clearly met the threshold for harassment and should be removed under Meta’s own rules.
Calling for stronger and more transparent responses, the quasi‑independent board, set up in 2020 and whose decisions are binding, issued nine policy recommendations. It urged Meta to expand “high risk AI” labels, add warning screens, have algorithms demote deceptive content, increase penalties for repeat sharers, and publish data showing when labels are applied. Co‑chair Pamela San Martin warned that deepfakes are increasingly used to silence women in public life.
Comprehension
- Who ordered Meta to remove the deepfake videos?
- What did the Scotland clip falsely show the councillor doing?
- Who was targeted in the second AI video, and what was her cause?
- Name one step the board urged Meta to add for high‑risk AI content.
- How did the board assess Meta’s safeguards overall?
Grammar focus
Adverbs and adverbial phrases: degree
Adverbs of degree tell us how strong or to what extent something is (place them before adjectives or other adverbs). Common words: very, much, far, clearly, consistently. (Examples constructed.)
- Observers said the fake clip was **extremely harmful**.
- Campaigners called the video **far more damaging** to reputations.
Grammar exercise
Choose the correct option in each sentence.
- Deepfakes can be (much / very) stronger than rumours.
- The board urged labels be (much / more) widely used.
- Versions of the clip were (far / highly) more viral.
- Review found safeguards (consistent / consistently) inadequate.
- The audio was not (very / much) synchronised with movements.
Discussion
- Is labelling deepfakes enough, or should removal be the default?
- Who should decide when algorithms demote content, and how?
- What data would real transparency about labels include?
- What support should targets of online harassment receive?
Vocabulary, reading, and exercises with instant feedback