Safety

OpenAI adds Google's SynthID to GPT-image-2 and launches public verifier

OpenAI announced it is integrating Google’s SynthID watermarking into GPT-image-2 and has released a public verification tool at openai.com/research/verify.

OpenAI adds Google's SynthID to GPT-image-2 and launches public verifier

OpenAI announced it is integrating Google’s SynthID watermarking into GPT-image-2 and has made a public verification tool available at openai.com/research/verify. The web tool accepts any image and returns a verdict in seconds on whether the image was generated.

How the technology works

The announced approach combines two elements:

  • C2PA: an open provenance standard that embeds a cryptographically signed manifest into a file; C2PA is already used by Adobe, Microsoft, Sony, Nikon, Leica, and Google.
  • SynthID: Google’s method that disperses an invisible signature across the image itself, designed to survive cropping, filtering, compression, and format conversion.

According to OpenAI, the verifier can identify generated content even after common manipulations such as screenshots, partial crops, or format changes, and provides a result within seconds.

Why this matters

Combining a file-embedded, signed manifest with a pixel-level, resilient signature aims to reduce the asymmetry between the ease of producing fake images and the difficulty of verifying them. Previously, the watermarking landscape resembled a standoff: Google had SynthID, OpenAI had C2PA-based pipelines, and the systems were not interoperable, which made widespread verification harder and increased the risk of untrusted image content online.

Practical effects and implications

  • The public verifier lowers the barrier to provenance checks: anyone can upload an image to the web interface and receive a quick determination.
  • The combined architecture could be a turning point for image trust online: if origins can be consistently attested, it reduces the incentive and value of fabricating trends or deceptive visual content.

Summary

OpenAI and Google aligned their watermarking approaches not as the result of regulatory compulsion but out of commercial self-interest, according to the announcement. By pairing C2PA manifests with SynthID’s resilient pixel signatures, the companies aim to make AI-generated images more reliably identifiable despite routine edits and conversions.