The rapid development of artificial intelligence is causing fake photos, deepfaked videos and misleading content to spread widely on social networks.
In that context, OpenAI has just announced a free AI image authentication tool to help users check the origin of images created by AI.
OpenAI has partnered with Google to integrate SynthID invisible copyright stamping technology into all images created by the company's AI products.
At the same time, the company also introduced a public verification tool, allowing users to upload photos to check if those images were created by AI.
OpenAI's new tool works based on two signal layers including SynthID and the C2PA open standard.
In which, C2PA adds direct recognition data to the metadata part of the image, while SynthID acts as an invisible copyright mark embedded in the image.
According to OpenAI, combining the two technologies helps increase reliability in content authentication.
Copyright marks can exist through many conversions such as screenshots, while metadata provides more detailed information. Two combined layers make authentication more reliable than just using one method," OpenAI said.
Currently, this tool mainly supports image detection created by OpenAI platforms such as ChatGPT, OpenAI API or Codex. However, the company said it is researching to expand compatibility with other AI tools in the future.
OpenAI's move takes place in the context of increasingly sophisticated and widespread AI image creation technology. With just a few simple steps, users can create images that are close to real, making distinguishing real and fake photos more difficult than ever.
However, many experts believe that current measures are still not enough to completely prevent deepfaking, especially content created from unreliable AI tools or intentionally avoiding authentication mechanisms.
To use the new verification tool, users only need to upload an image in PNG, JPG or WEBP format. The system will analyze and notify whether the image contains the C2PA signal, SynthID mark or not.
OpenAI recommends cutting close to the main content of the image and avoiding uploading files containing many images to increase accuracy.
However, during testing with an image created using OpenAI's own Images 2.0 model, the system responded that "no evidence was found that the content was created using OpenAI tools". This shows that AI authentication technology still has certain limitations.
