Finally uniform AI labeling? Joint effort by OpenAI and Google
Google and OpenAI intend to mark their AI content with the same watermark and metadata standards in the future. This gives cause for hope.
(Image: Jack_the_sparow/Shutterstock.com)
Both Google and OpenAI intend to mark their AI content with the SynthID watermark and the Content Credentials metadata standard in the future. The fact that two major AI providers are now relying on the same system could mean progress towards an industry-wide standard.
For several years, the industry has been searching for ways to clearly mark the origin of media content. Especially since AI has become available to everyone, it is becoming increasingly difficult to recognize whether a text, photo, video, or audio file was created by a human or a machine.
No uniform standards so far
Since 2019, various companies have been trying to create a uniform standard for proof of origin. Each media content should be accompanied by metadata with information about who created it, when, where, and with what tools. This includes camera manufacturers like Sony, media companies like the BBC, agencies like Reuters, and tech giants like Google, Meta, or Adobe, who have joined forces in the Coalition for Content Provenance and Authenticity (C2PA). However, there have been repeated unilateral actions and technical problems in the past. Nevertheless, C2PA's “Content Credentials” is currently the most recognized standard, which companies like OpenAI have also been using for various AI products for some time.
However, metadata can be easily altered or removed. Social media platforms like YouTube even delete them by default when uploading to the platform. What helps is linking the data with a watermark.
How watermarks can help
Companies like Digimarcs, which are also involved in the Content Credentials standard, have been coupling metadata and watermarks since 2023. The idea behind it is: While the former can be easily removed, watermarks are “burned” into the medium. Depending on the media category, different technologies are used; for images, for example, there is the possibility of incorporating a kind of image noise made of pixels. These are then altered in a specific pattern. The human eye cannot see this alteration, but machines can read it. If an image now has a watermark but no metadata, the original metadata can potentially be restored thanks to the unique watermark identifier. And even if not, the user at least recognizes that the image might be suspicious.
With SynthID, Google wants to set a new standard for watermarks, and OpenAI, as well as Kakao and ElevenLabs, have announced that they will join. At OpenAI, the technology is currently only available for images created with ChatGPT, Codex, or the OpenAI API. At Google, marking with SynthID is already standard for many of its AI services, and more and more applications are also adding C2PA Content Credentials metadata.
Starting August 2026, the EU AI Act will mandate that AI content must be clearly marked by all providers offering AI products in Europe. Otherwise, there will be penalties of up to 15 million euros, or 3 percent of annual revenue.
Videos by heise
Same proofs of origin for everything
The latest Google Pixel phones, like Samsung's recent Galaxy class, are also supposed to mark every image and soon video recording with a proof of origin according to the Content Credentials standard, just as some cameras have been doing for a longer time. This is intended to make not only fake but also real images clearly identifiable. Instagram also announced that it will soon be adding metadata to camera recordings on its platform. And the photo agency Getty will now be required to use content credentials. The idea: Anything without metadata or watermarks should be suspicious in the future.
However, there is still no uniform solution for how users can read this AI identifier. Instead, users have to laboriously upload content to various tools because many providers can only read the proof of origin from their AI.
On Tuesday, Google announced that the SynthID watermark can now be checked not only in the Gemini app but also in Search and soon in Chrome for photos. Other media types are to follow. The prerequisite is having a Google account. However, Google is shutting down its previous, registration-free verification portal, thus forcing users to register with its services. In return, Google will now output not only the presence of a watermark but also the metadata of Content Credentials – for this, users previously had to access a separate website. This has been possible in the Gemini app since this week, and this function is expected to come to Search and Chrome soon.
Empfohlener redaktioneller Inhalt
Mit Ihrer Zustimmung wird hier ein externes Video (TargetVideo GmbH) geladen.
Ich bin damit einverstanden, dass mir externe Inhalte angezeigt werden. Damit können personenbezogene Daten an Drittplattformen (TargetVideo GmbH) übermittelt werden. Mehr dazu in unserer Datenschutzerklärung.
OpenAI, on the other hand, plans to launch a separate verification page soon, which will initially only be able to recognize OpenAI content and later also AI content from other providers.
Other providers, such as the Verify page of the Content Authenticity Initiative and the Chrome extension from Digimarc, are options for reading existing proofs of origin – and for a fee, also for watermarking and adding metadata to your content. Anyone who wants to publish a photo or video can thus ensure that its origin is clearly recognizable to everyone.
(vbr)
