Google and OpenAI bet on invisible watermarks to curb AI fakery

Google is moving its SynthID watermarking technology out of the Gemini app and directly into Chrome and Search, aiming to normalize the verification of AI-generated files. The update allows users to detect both SynthID markers and C2PA provenance metadata—an industry standard for tracking media origins—from a single interface. By embedding these tools into the browser, Google hopes to bypass the limitations of platforms that currently strip away metadata or fail to display it entirely.

OpenAI is also expanding its commitment by embedding SynthID into images generated via ChatGPT and its API, effectively doubling down on the technology. This move arrives alongside Meta’s plan to use C2PA metadata to label camera-captured images on Instagram, a shift Instagram head Adam Mosseri believes is necessary as users lose the ability to assume digital content is authentic by default.

Despite the collaborative push, the systems face significant practical hurdles. C2PA metadata is notoriously fragile, often discarded when users take screenshots or upload files to social media. While SynthID is considered more robust against tampering, both technologies rely entirely on voluntary adoption by model creators. As it stands, the open-source models most frequently used for malicious deepfakes remain largely outside this ecosystem. Google’s latest expansion serves as a high-stakes test: it is now forced to demonstrate whether its dual-standard approach can actually filter online misinformation or if it remains a superficial fix to a systemic problem.