Watermarking, a strategy to identify AI-generated images and text, is proving to be unreliable, according to a study by University of Maryland computer science professor Soheil Feizi. The research revealed that bad actors can easily evade or manipulate watermarks, including those invisible to the naked eye. Despite these findings, tech giants continue to develop watermarking technology to combat misinformation, while others believe it can be part of the solution if used in combination with other technologies.
Read more at Ars Technica…
