It’s make or break time for AI labeling systems

We’re about to find out whether systems designed to make it easier to detect deepfake and AI-generated content can actually eliminate them. SynthID and C2PA Content Credentials, two breakthrough technologies for invisible tagging of image, video and audio files with information about their origins, are getting their biggest expansion yet, and with it the opportunity to turn the tide against unnamed AI counterfeiting that deceives people online.

Yesterday during I/O, Google announced that the ability to check whether images have SynthID tags — the invisible watermarking system applied to content generated by Google AI models — is available. Coming to Chrome and searching. This is important because Chrome definitely dominates Global market share of web browsers and search engines, so AI verification tools are being paid more attention. It also simplifies the inspection process; If you currently want to check an image for SynthID tags, you are expected to upload it to the Gemini app.

Not only that, but Google Verified interfaces will now do this also Check whether these files contain C2PA information – source metadata embedded in the content at the point of creation to tell us how it was created or processed and whether AI tools were used during the process. This C2PA certification allows users to verify suspicious images from a single interface instead of hopping between the Gemini app and dedicated C2PA verification portals since files may have only one type of label or none.

This is the kind of collaborative effort we’ve been waiting for. While the two systems work differently, both Google and the Content Authentication Initiative (which exists to promote C2PA standards) have made similar claims about what is required for them to work: for everyone to be on board. This means that more AI models need to include this data, and the online platforms where AI fakeouts are most often shared need to clearly display that information. For the latter, having validators built into the web browser can serve as a workaround on websites that do not verify or provide AI metadata to their users.

OpenAI is also participating in this expansion, announcing yesterday that it will now do so Include SynthID In images generated by ChatGPT, Codex, and OpenAI API. The company does include C2PA metadata in the content generated, but I found that This is often abstracted When published on other platforms. OpenAI itself has also wanted to temper expectations around C2PA, despite being a steering member of C2PA and now confirming its commitment to the standard. This is what OpenAI said C2PA help pagebefore it was Updated to include SynthID yesterday:

“Metadata like C2PA is not a silver bullet to address provenance issues. 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 remove it as well. Therefore, an image lacking this metadata may or may not be created using ChatGPT or our API.”

For something considered the absolute best in content authenticity technology, this seems incredibly flimsy. Even Google describes C2PA as… the Industry standard, it’s being Presented to global governments As a solution to satisfy transparency and labeling requirements in AI. But although it is increasingly It is certified by AI, hardware and software providersI’ve rarely seen it used successfully to check fake AI in the wild. SynthID seems more powerful by comparison because it can’t be easily removed – given how limited its reach is compared to C2PA, I can recall several instances where fact-checkers and media agencies have pointed to its use in debunking online deepfakes.

C2PA and SynthID can work collaboratively to create a broader security net. This isn’t an industry that would benefit from a verification standards war, but Google has a clear opportunity here to prove whether its system is more reliable and attract some of the spotlight that C2PA has given itself. To prevent this from happening, C2PA needs to prove that it can actually be used to demystify the source of the content we see online.

Such an opportunity has already presented itself: Yesterday Google announced that Meta will start using C2PA metadata to tag photos on Instagram captured by the camera. Meta didn’t respond to our questions about what this will look like or what cameras will be supported, though I assume it will include labels that say something like “Taken on Pixel 10,” similar to the “Sent from my iPhone” notes applied to emails. This would effectively help Instagram users distinguish between “real” photos and convincing AI-driven fakes, which play a role in the future predicted by the Instagram boss. Adam Mosseri Regarding the need to move away “from assuming that what we see is hypothetically real.”

If tagging works, that is. Instagram already scans photos for C2PA information, and its attempts to categorize AI-generated content have subsequently thrown the platform into disarray. Apply artificial intelligence labels to images The photographers insisted that they took the pictures themselves.

I wouldn’t be too quick to praise Google for this team either. The company preaches the importance of AI transparency and combating digital deepfakes, while developing the same technology that is being used to mislead people. It has positioned itself as the resource and solution. I can forgive that if SynthID makes a noticeable difference in the fight against deepfakes, but I’m not getting my hopes up given the scale of the problem.

Whether SynthID and C2PA are robust or not, they can’t detect watermarks unless they’re added in the first place, and I doubt many of the open source models used to create truly outrageous deepfake content are lining up to adopt these systems. Sourcing has never been a perfect solution, but now Google and C2PA have the opportunity to prove that it’s not a waste of time.