As fears mount over the proliferation of deepfake images that could influence this year’s many elections, OpenAI today announced the creation of a new tool to detect if an image was made with its own DALL-E AI image generator.

Unfortunately, OpenAI’s deepfake detector only works successfully on images made by its own generative artificial intelligence, but it’s a start. For years now, deepfake images have plagued the internet, especially manipulated sexual images. The worry is that as image generators improve, the web, especially social media, will be inundated with offensive fakes.

OpenAI has said it’s under no illusions that deepfakes won’t become a bigger problem as technology improves, which is why it’s so important to create detection tools. With its new detector, AI predicts if an image has been altered from its original state. The company said it had over a 98% accuracy rating in detecting images made with DALL-E. Still, accuracy slumped to just 5% to 10% with images made by other companies’ image generators.

Progress will be slow, which is why OpenAI is now part of the steering committee for the Coalition for Content Provenance and Authenticity, or C2PA, which joins other firms in advancing technical standards for digital content provenance. Google LLC recently signed on to the board, joining Adobe Inc., Intel Corp., Meta Platforms Inc., Microsoft Corp. OpenAI and Sony Corp.

The C2PA standard will add a “nutritional label” to content that has been altered, whether it is a still image, text, video or audio. The label should say when the image was altered and how it was altered, which may ameliorate the fears of politicians who are calling for more legislation around AI. The EU and the UK appear to be ahead of the U.S. in this regard.

“Our efforts around provenance are just one part of a broader industry effort – many of our peer research labs and generative AI companies are also advancing research in this area,” said OpenAI in a blog post about its new detector. “We commend these endeavors — the industry must collaborate and share insights to enhance our understanding and continue to promote transparency online.”

Image: Mariia Shalabaieva/Unsplash