Oversight Board blasts 'inadequate' Meta safeguards for AI deepfakes

Meta on Thursday moved to remove AI-generated videos targeting two women, including a British politician, after the tech giant's Oversight Board blasted its "fundamentally inadequate" safeguards against proliferating deepfakes.

The Oversight Board, the quasi-independent body the company funds to review its moderation decisions, earlier Thursday demanded that Meta remove the fabricated videos from its platforms, saying the cases undermined its own policies on hate speech, misinformation and harassment.

The first AI-generated video depicted a Labor Party councilor in Scotland making incendiary comments about refugees.

There was no public record of the politician making the comment and the video showed telltale signs of fabrication, including audio that was not synchronized to her facial movements.

The second fabricated video targeted a Muslim woman campaigner in Europe and sought to discredit her role as a health advocate by falsely depicting her consuming junk food while giving advice about mindful eating and engaging in absurd exercises.

The board said Meta's policies are "consistently and fundamentally inadequate" to address the growing flood of AI-generated content on its platforms, which include Facebook and Instagram.

"From politicians to private citizens, AI-generated deepfakes are increasingly being used to harass and silence women from engaging in public discourse," Oversight Board Cochair Pamela San Martin said in a statement.

"These cases demonstrate a broader, troubling pattern in which women who engage publicly on issues are disproportionately subjected to harassment and misinformation." The Oversight Board's decisions on individual cases are binding on Meta, but its broader policy recommendations are not.

Meta said it welcomed the board's decision in both cases and will comply within seven days.

"Meta and other social media platforms need more robust policies to address the proliferation of deepfakes, including a much stronger approach to labeling deceptive content, stricter penalties for repeat offenders, and the removal of deepfakes harassing private individuals," said San Martin.

Last week, the board urged Meta not to replace professional fact-checking with Community Notes, a crowdsourced model for debunking falsehoods popularized by billionaire Elon Musk's platform X.

The plea came after Meta said it had begun testing Community Notes in 16 Spanish-speaking countries in Latin America, in a step toward replacing its fact-checking program.