DEVOURED
OpenAI is Making it Easier to Check if an Image was Made by Their Models
OpenAI announced on May 19, 2026, new measures including C2PA metadata and Google's SynthID invisible watermarks to help identify images generated by its AI models.
What: On May 19, 2026, OpenAI introduced two methods for identifying AI-generated images: adopting the C2PA open standard for metadata signals and integrating Google's SynthID invisible watermarks. They are also previewing a public verification tool to check for these signals, though these measures currently only apply to images from OpenAI's own products.
Why it matters: This move by OpenAI, in partnership with Google, signals a growing industry effort to combat misinformation and deepfakes by establishing content provenance, though its effectiveness is limited by broader industry adoption.
Deep dive
- OpenAI's new strategy to combat AI-generated image misinformation involves two complementary approaches.* The first is the adoption of C2PA, a visible metadata signal standard established by the Coalition for Content Provenance and Authenticity (founded 2021).* C2PA signals are accessible in file metadata, making them useful among trusted parties but susceptible to manipulation.* The second is a partnership with Google to embed SynthID, an invisible and more robust watermark designed to persist even after transformations like screenshots or resizing.* OpenAI is also launching a public verification tool that can check for both C2PA and SynthID signals.* These protections are currently limited to images generated by OpenAI's own models, not the wider ecosystem of AI tools.* The dual approach aims to combine the detailed information provided by metadata with the durability of invisible watermarking.
Decoder
- C2PA (Coalition for Content Provenance and Authenticity): An open technical standard that adds cryptographic metadata to digital content, signaling its origin and any modifications, to help establish trust and combat misinformation.* SynthID: An invisible, imperceptible watermark developed by Google that is embedded directly into the pixel data of AI-generated images, designed to remain detectable even after image modifications or compression.* Deepfake: A portmanteau of 'deep learning' and 'fake,' referring to synthetic media in which a person in an existing image or video is replaced with someone else's likeness using AI techniques.
Original article
Full article content is not available for inline reading.
Explore all 49 articles in today's digest
Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi