AI deepfakes are cheap, easy, and coming for the 2024 election

Watermarks

Watermarks are being explored as a solution to identify AI-generated media, but they face significant challenges. Adi Robertson explains that while companies are adopting initiatives like C2PA, which includes watermarking systems, the effectiveness is limited by user engagement and technical vulnerabilities 1. She notes that removing watermarks can be as simple as using a screenshot tool, making it difficult to ensure authenticity 2.

It's very, very hard to create a perfect watermarking system.

--- Adi Robertson

The goal is to train users to expect authenticity markers, but widespread adoption and enforcement remain hurdles.

   

Industry Standards

The development of industry standards like C2PA aims to establish content provenance, yet challenges persist. Nilay Patel discusses the difficulty in enforcing watermark requirements, as open-source tools can easily bypass them 3. He highlights the ongoing cat-and-mouse game between AI developers and those trying to regulate content creation.

The real problems are in distribution.

--- Nilay Patel

The focus is shifting towards distribution, where platforms are pressured to label and moderate deepfakes, though their commitment varies 4.