Unpacking OpenAI and Google's Latest Moves in Content Provenance

While Google and OpenAI’s adoption of C2PA is a critical milestone for digital trust, building a truly authentic information ecosystem requires bridging access gaps, supporting diverse verification me

Jun 16, 2026

Two tech titans have taken up armor in the fight for the future of digital trust. Earlier in May 2026, OpenAI and Google both made significant announcements about their adoption of content provenance methods, including one that Starling Lab helped to pioneer.

OpenAI is adopting the C2PA standard, adding SynthID watermarking via Google DeepMind for images generated in ChatGPT, Codex, and the OpenAI API. It will also preview a public verification tool that checks for Content Credentials and SynthID signals in OpenAI-generated images.

Meanwhile, Google is expanding consumer-facing verification in Gemini, Search, and Chrome, adding C2PA Content Credentials checks, and launching a Google Cloud AI Content Detection API for businesses; it is also extending Pixel’s Content Credentials support to video on Pixel 8, 9, and 10 devices.

We’ve long argued that proving authenticity requires a “defense in depth” approach. These announcements and increased adoption are important milestones for the standards and technologies.

At the same time, it’s worth noting that creating a trustworthy – and equal – digital information environment will take more than being able to deduce what was generated synthetically by Google or OpenAI.

We think that furthering trust in online media also must consider:

Giving authentic content its own provenance signals: Other camera and smartphone manufacturers should work on incorporating means of digitally attesting to the real content their devices capture, following Google’s additions to the Pixel family.
Thinking beyond technical data: The Creator Assertions Working Group (CAWG) provides a blueprint for creating provenance records supporting individuals and publishers in creating human-generated metadata and expression of their creative rights.
Bridging a problematic access gap: The tech isn’t readily accessible to all. Fortunately, we can promote open tools like Proofmode that bring the latest in provenance tech to all gear.
Maintaining (some) signals as they travel on- and offline: Social media and messaging apps, content delivery networks, web browsers and operating systems, should consider how to preserve markers of trust associated with content wherever it is being consumed.
Supporting after-the-fact assertions and endorsements: We think there’s tremendous value in a pluralistic system of verification based on corroborating claims made about media assets, and in public forums e.g. social media. See our case study and prototype looking at Chris Morris’ fascinating Panama invasion archive.

It’s also worth considering the wider political context around these decisions. Although OpenAI and Google make no mention of it, the EU AI Act and California AI Transparency Act require all large commercial gen AI offerings to take precisely these steps before August of this year, and to embed latent disclosures into generated materials and provide tools for checking if materials are synthetic. The California legislation goes further and requires camera manufacturers to likewise embed trust signals by 2027.
For more on these initiatives, see our Dispatch on the California AI Transparency Act, as well as our international tracker of relevant initiatives.

Starling Lab

Discussion about this post

Ready for more?