We’re about to find out if systems designed to make deepfakes and easy-to-spot AI-generated content really work. SynthID and C2PA content information, two different technologies for anonymously tagging image, video, and audio files with information on their origin, are getting the biggest increase to date, and with it, an opportunity to change the situation against the unwritten AI deception that deceives people on the Internet.
Yesterday during its I/O conference, Google announced that the ability to verify that images carry SynthID tags – an invisible watermarking system used in content generated by Google AI models – is coming to Chrome and Search. That’s important because it’s completely Chrome he rules global market share for web browsers and search engines, so AI validation tools are thrust in front of many eyes. It also simplifies the testing process; if you now want to check the SynthID tag image, you are expected to upload it to the Gemini app.
Not only that, but Google verification links will now again check whether these files contain C2PA information — origin metadata embedded in the content during creation to tell us how it was created or used and whether AI tools were used during the process. This C2PA adoption allows users to check suspicious images from a single interface instead of jumping between the Gemini app and dedicated C2PA verification sites as files may have only one type of label or none at all.
This is the kind of collaborative effort we’ve been waiting for. Although both systems work differently, both Google and the Content Authenticity Initiative (which exists to promote the C2PA standard) have made similar claims about what is needed for them to work: for everyone to be on board. That means many AI models need to embed this data, and online forums where AI fakery is often shared need to clearly display that information. Finally, having authentication tools built into the web browser can serve as a workaround for websites that do not inspect or present AI metadata to their users.
OpenAI is also getting involved in this expansion, announcing yesterday that it will now embed SynthID in images generated by ChatGPT, Codex, and the OpenAI API. The company already includes C2PA metadata in generated content, but I’ve found that this is often stripped when posting to other platforms. OpenAI itself has also sought to lower expectations regarding C2PA, despite being a guiding member of C2PA and now reaffirming its commitment to the standard. Here’s what OpenAI said on its C2PA help page, before it was updated to include SynthID yesterday:
“Metadata like C2PA is not a silver bullet to deal with the problems of its appearance. It can easily be removed accidentally or on purpose. For example, many social networks today remove metadata from uploaded images, and actions like taking a screenshot can also remove it. Therefore, an image without this metadata may or may not have been generated through ChatGPT or our API.”
For something that’s considered the best technology for content authenticity, that sounds incredibly dull. Even Google defines C2PA as i industry standard, and is being sent to international governments as a solution to appease AI transparency and labeling requirements. But despite its increasing adoption by AI, hardware, and software providers, I rarely see it being used successfully to verify AI fraud in the wild. SynthID seems more robust in comparison because it can’t be cracked easily – because of how limited its reach is compared to C2PA, I can recall several times when fact-checkers and media organizations have cited its use in mitigating deepfakes online.
Both C2PA and SynthID can work together to spread a wider safety net. This is not an industry that can benefit from a war of authentication standards, but Google has a clear opportunity here to prove whether its system is more reliable and use some of the shine that C2PA has created for itself. To prevent this from happening, C2PA needs to prove that it can actually be used to determine where the content we see online is coming from.
Such an opportunity has already presented itself: Google announced yesterday that Meta will start using C2PA metadata to tag photos on Instagram taken by the camera. Meta hasn’t responded to our questions about what this will look like or what cameras will be supported, though I imagine it will include labels that say something like “shot on Pixel 10,” as in the “sent from my iPhone” notes used in emails. This will effectively help Instagram users to distinguish “real” photos from deceptive AI convincing, which plays into the future predicted by the head of Instagram Adam Mosseri about the need to move away from “thinking that what we see is automatically real.”
If the labeling works, of course. Instagram is already testing photos for C2PA information, and its efforts to label AI-generated content have landed the platform in hot water after it added AI labels to photos that photographers insist they take themselves.
I won’t be too quick to praise Google for this team. The company preaches the importance of AI transparency and the fight against digital deepfakes, all the while developing the technology used to mislead people. It is positioned as both a supplier and a solution. I can justify that if SynthID makes a noticeable difference in the fight against deepfakes, but I don’t have my hopes up given the magnitude of the issue.
Strict or not, SynthID and C2PA can only detect watermarks if they are added in the first place, and I doubt that many of the open source models used to produce deplorable content are on their way to using these systems. Provenance will never be a perfect solution, but now Google and C2PA have a chance to prove that it’s not a complete waste of time.
