OpenAI teams up with rival Google: ChatGPT fully integrates DeepMind’s invisible watermark and launches an image true-or-fake verification tool

AIMPACT News, May 20 (UTC+8), according to Beating monitoring, OpenAI announced a collaboration with former rival Google DeepMind, with full integration of SynthID invisible watermarking into images generated by ChatGPT, Codex, and the API, and it has formally become a C2PA image credential standard-compliant product while launching a public preview of an image authenticity verification tool.

As an industry-standard C2PA metadata system, it relies on cryptographic signatures to record provenance, but when users download images, take screenshots, or convert image formats, this metadata is highly likely to be lost. To patch this vulnerability, OpenAI introduced SynthID invisible watermarking technology developed by Google DeepMind, embedding identifiers that are imperceptible to the naked eye directly into the image pixels. Even if the image undergoes secondary processing such as screenshots, SynthID can still remain. The dual-layer defense mechanism combines the rich context provided by metadata with the persistence of invisible watermarking.

Along with this update, OpenAI also launched an authenticity verification website (openai.com/research/verify), and it is currently open to the public for a preview. Users can upload images directly on the web, and the tool will detect whether the images contain OpenAI’s Content Credentials credentials or SynthID invisible watermarking. However, OpenAI is extremely cautious in its detection conclusions. Because the watermarking and metadata may still be forcibly stripped in extreme cases, even if the tool does not find any anti-counterfeiting signals, it cannot decisively conclude that the image was definitely not generated by AI.

(Source: BlockBeats)