OpenAI is moving to make its AI-generated images easier to spot just as synthetic visuals become nearly indistinguishable from reality, escalating the race between detection tools and ever-more-realistic fakes.

Early groundwork: C2PA and SynthID emerge

Over the past several years, Google developed SynthID, a hidden watermarking system it says has already been used to label “100 billion images and videos, plus 60,000 years’ worth of audio,” and designed it to survive compression, cropping, and rotation.¹ At the same time, the Coalition for Content Provenance and Authenticity (C2PA), founded in 2021, pushed an open standard that tags files with metadata describing how they were created.²

Google committed to both strategies, embedding C2PA metadata in Pixel photos and videos and integrating SynthID directly into the pixels and waveforms of AI outputs.¹

May 19: OpenAI announces dual-layer provenance

On May 19, OpenAI said that images generated by ChatGPT, its API, and related tools will now carry Google’s SynthID watermark alongside C2PA content credentials, calling it a “multi-layered approach” to AI labeling.³ The company emphasized that “watermarking can be more durable through transformations like screenshots, while metadata can provide more information than a watermark alone. Together, they make provenance more resilient than either layer would be on its own.”⁴

The move includes a public verification portal that lets anyone upload an image to check for C2PA and SynthID signals and see if it was generated by OpenAI models.³ TechCrunch notes the protections “only apply to images generated by OpenAI products,” meaning they won’t touch the “flood of imagery coming from less reputable AI tools.”²

Industry and public perspectives

Coverage in The Verge framed the update as OpenAI “getting serious about AI detection and labeling,” arguing that combining C2PA and SynthID could make its content “less likely to slip through gaps in verification systems.”³ Google’s broader rollout of SynthID to firms like OpenAI and Nvidia was described as a significant, if not foolproof, step toward telling “truth from fiction.”¹

OpenAI co-founder Greg Brockman promoted the feature on X, writing, “SynthID for checking if an image was generated by OpenAI,” while amplifying OpenAI’s pledge to add “new ways for people to identify AI-generated images and understand where they came from” via C2PA, SynthID, and a public verification tool.Greg Brockman (gdb)⁵ Meanwhile, a viral post reshared by Elon Musk highlighted that systems like Grok Imagine are already producing imagery whose “realism is insanely good… blurring the line between AI generation and reality,” underscoring why provenance efforts are urgently needed.Elon Musk (elonmusk)⁶