Here's How Big Tech Is Tackling Deepfakes

Here’s an increasingly familiar feeling: You see an image and you can’t tell if it was made or manipulated using artificial intelligence. It’s a big problem, given the potential of fake images to influence elections, among other things. Fortunately, many of the companies whose technology makes it easy to create and spread AI-generated content now are introducing ways to identify and label that content. 

It may be a while before this problem is really solved, however. No single standard has emerged yet, and it isn’t clear how effective any of them will be. Take it from the companies themselves. Last month, OpenAI said it would embed metadata into images generated in ChatGPT and its application programming interface, to make it easier to detect them. The company also said it would embed metadata into videos made using Sora, if it deployed the video generation model in a product. But OpenAI added that the move “is not a silver bullet to address issues of provenance” and markers “can easily be removed either accidentally or intentionally”—for example, by screenshotting an image.