Not to be outdone by a rival, OpenAI today announced it is updating its marquee app ChatGPT and the AI image generator model integrated within it, DALL-E 3, to include new metadata tagging that will allow the company, and theoretically any user or other organization across the web, to identify the imagery as having been made with AI tools.
The move came just hours after Meta announced a similar measure to label AI images generated through its separate AI image generator Imagine and available on Instagram, Facebook, and Threads (and, also, trained on user-submitted imagery from some of those social platforms).
“Images generated in ChatGPT and our API now include metadata using C2PA specifications,” OpenAI posted on the social platforms X and LinkedIn from its corporate account. “This allows anyone (including social platforms and content distributors) to see that an image was generated by our products.”
OpenAI said the change was in effect for the web right now, and would be implemented for all mobile ChatGPT users by February 12.
VB Event
The AI Impact Tour – NYC
We’ll be in New York on February 29 in partnership with Microsoft to discuss how to balance risks and rewards of AI applications. Request an invite to the exclusive event below.
Request an invite
The company also included a link to a website called Content Credentials where users can upload an image to verify if it is AI-generated or not, thanks to the new code it is applying. However, the change is only in effect for newly generated AI images with ChatGPT and DALL-E 3 — all the ones generated before today won’t have the metadata included in them.
What is C2PA?
The Coalition for Content Provenance and Authenticity, or C2PA, is a relatively new effort that sprang from the Joint Development Foundation, a non-profit made up of several other organizations that are ultimately funded by the likes of Adobe, ARM, Intel, Microsoft (OpenAI’s investor and business partner), The New York Times (currently suing OpenAI for copyright infringement), the BBC, CBC, and several more media and tech firms.
It was founded back in February 2021, before ChatGPT was even released, with the goal of “developing technical standards for certifying the source and history or provenance of media content,” to “address the prevalence of disinformation, misinformation and online content fraud.”
In January 2022, the C2PA released its first technical standards for how developers at responsible AI model makers and companies can code in metadata — extra data not vital to the image itself — that will reveal under some circumstances that it was created by an AI tool.
That mission has taken on renewed urgency as of late, with high-profile examples such as AI-generated explicit and nonconsensual deepfakes of Grammy Award-winning musician Taylor Swift spreading widely on the social platform X, as well as similarly nonconsensual explicit deepfakes of high school students among their peers.
Separately but relatedly, AI video and voice cloning were blamed for a scam in which a Hong Kong, China-based employee was tricked into transferring $25 million to scammers from an unnamed multinational company, and already, voice cloning is being used to influence the U.S. 2024 election cycle.
OpenAI earlier this year said it would introduce C2PA to combat disinformation ahead of the 2024 global elections taking place, and today’s news appears to be the company making good on that promise.
C2PA seeks to help platforms identify AI-generated content by embedding metadata in the form of an electronic “signature” in the actual code that makes up an AI image file, as shown in an example posted by OpenAI on its help site.
However, OpenAI readily admits on its help site that: “Metadata like C2PA is not a silver bullet to address issues of provenance. It can easily be removed either accidentally or intentionally. For example, most social media platforms today remove metadata from uploaded images, and actions like taking a screenshot can also remove it. Therefore, an image lacking this metadata may or may not have been generated with ChatGPT or our API.“
Furthermore, this metadata is not immediately visible to a casual observer — instead, they must expand or open the file description to see it.
Meta, by contrast, showed off a preview of its platform-wide AI labeling scheme earlier today which would be public-facing and include a sparkles emoji as an immediate signifier to any viewer that an image was made with AI tools.
However, it said the feature would not begin rolling out until “the coming months” and was still being designed. It too, relies on C2PA as well as another standard called the IPTC Photo Metadata Standard from the International Press Telecommunications Council (IPTC).
