The Issue at Hand
Meta’s Oversight Board has called out the company’s inadequate response to AI-generated explicit images of female public figures on its platforms. The board’s investigation, spanning three months, focused on two high-profile cases involving deepfake nude images of public figures from India and the U.S. While both images violated Meta’s rules against “derogatory sexualized photoshop” content, the company’s handling of these cases revealed significant shortcomings in its policies and enforcement mechanisms.
Key Findings and Recommendations
- Inconsistent enforcement: The image of the Indian public figure remained viewable due to automated closures of reports and appeals, while the image of the American celebrity was promptly removed and added to Meta’s automated enforcement system.
- Language update needed: The board suggested moving the prohibition on “derogatory sexualized photoshop” to the Adult Sexual Exploitation Community Standard and replacing “derogatory” with “non-consensual” and “Photoshop” with a more general term for manipulated media.
- Policy expansion: Recommendations include specifying that AI-generated or manipulated images should be covered by non-consensual content policies, regardless of commercial intent or private setting production.
Broader Implications
The board’s findings underscore the urgent need for tech companies to adapt their policies to the rapidly evolving landscape of AI-generated content. The prevalence of deepfake sexually-explicit images targeting women (99% of cases) highlights the gendered nature of this issue. The board also noted the potential for severe consequences in socially conservative communities, citing a tragic incident in Pakistan. By clarifying rules and emphasizing the lack of consent, Meta could potentially reduce the proliferation of AI deepfakes on its platforms, addressing a growing concern in the digital age.
