Google is trying to make deepfake detection more accessible for everyone

Google is expanding AI detection capabilities to Chrome and Search, with the aim of making it easier for people to identify deepfakes. The updates, announced at Google I/O today, cover not only SynthID — the invisible watermarking technology developed by Google DeepMind — but also content embedded with C2PA content credentials, making both systems more accessible for users to learn how the content they see online was made or manipulated.

To start, Google says verification for images that carry SynthID markers (which indicate they’ve been made with Google’s AI tools) is coming to Search features starting today, including Google Lens, AI Mode, and Circle to Search. Uploading or selecting online images with these tools can tell users more about where the image originated, with support for Chrome expected to roll out “in the coming months.” Google says the Chrome expansion is also powered by Gemini, and will work by selecting or circling an image on the browser page. The image will then appear on the right-side panel, where users can ask “Is this made with AI?” or “Is this AI generated?”

SynthID verification for image, video and audio content is already available through the Gemini app. The expansion to Chrome and Search is image-only for now, but Google says it plans to expand to other types of content in the future.

As part of this update, Google’s verification interfaces will also check if content contains C2PA Content Credentials alongside its own SynthID markers. That means people can now check for both provenance systems from a single interface instead of jumping between Gemini and dedicated C2PA checker portals, creating a more streamlined process for deepfake detection. C2PA verification is rolling out to the Gemini app today, with Search and Chrome support to follow in the coming months.

This isn’t the only effort to expand C2PA across Google’s platforms — this metadata was already embedded on images taken using Pixel 10 phones, and now Pixel 8, 9 and 10 phones will start including it for video files in the coming weeks. Google also says that Meta will start labeling camera-captured media on Instagram using C2PA, though Meta hasn’t responded to our request for more information about this upcoming feature. Meanwhile, support for Google’s SynthID watermarks are coming to more AI content generated by OpenAI, Kakao, and ElevenLabs, though Google doesn’t specify what models or content this will apply to.

Rounding off these updates is a new AI content detection API for Google Cloud’s Gemini Enterprise Agent Platform, which aims to make it easier for organizations to identify AI-generated media. Google says this capability is launching with a group of unspecified trusted partners, and that the API will be fine tuned based on their feedback.

“This gives businesses a powerful way to spot AI content made by both Google and other popular models, helping them decide how to evaluate and manage media across their own platforms – whether that’s for backend operations like sorting feeds and preventing insurance fraud, or for user-facing content like fact-checking and labeling synthetic media,” Google said in its announcement blog.

Together, these expansions should help to make AI detection more accessible for both companies and everyday web users. The Chrome expansion is especially notable as it’s the most used browser across the globe — many online platforms don’t label AI content natively, so this provides an easier workaround compared to saving images and manually uploading them to dedicated C2PA and SynthID detection interfaces.

That said, Google DeepMind VP Pushmeet Kohli told The Verge that Google has scrapped plans to launch a dedicated verification portal for SynthID. For Google’s own systems at least, that means SynthID markers may only be identified on Gemini-powered platforms going forward, forcing people to use Google’s own AI tools to identify AI content generated by Google models. That may not go down well with folks who want to entirely avoid interacting with generative AI systems.