Facebook parent company, Meta, has unveiled changes to its policies regarding digitally created and altered media ahead of the upcoming U.S. elections, to regulate disinformation generated by advanced artificial intelligence (AI) technologies.
In a blog post, Meta’s vice president of content policy, Monika Bickert, outlined the new measures. Starting in May, Meta will introduce “Made with AI” labels for videos, images, and audio generated using AI, expanding its previous policy which only targeted a limited subset of manipulated videos.
Furthermore, Meta will implement separate and more prominent labels for digitally altered media that can pose a high risk of materially deceiving the public, regardless of whether AI or other tools were used in their creation.
Previously, Meta had announced plans to detect images produced using third-party generative AI tools by embedding invisible markers into the files, although no start date was specified at that time.
Reuters reported that a spokesperson for Meta confirmed that the new labelling approach will be applied to content across Meta’s platforms, including Facebook, Instagram, and Threads, with immediate effect. However, different rules govern other Meta services such as WhatsApp and Quest virtual reality headsets.
Tech researchers have already displayed concern about the potential impact of new generative AI technologies on the upcoming US electoral process. Political campaigns have already begun utilising AI tools in various contexts, raising questions about adherence to guidelines set by Meta and leading AI market player, OpenAI.
Earlier this year, Meta’s oversight board criticised the company’s existing rules on manipulated media as “incoherent,” following a review of a video posted on Facebook in which real footage of US President Joe Biden was altered to falsely suggest inappropriate behaviour.
The board has called for a broader application of the policy to cover non-AI content and audio-only content, stressing that such content can be equally misleading as AI-generated content.
