OpenAI announced two major initiatives on May 19, 2026, to address the rise of AI-generated deepfakes and misinformation. The company partnered with Google to integrate SynthID, an invisible watermarking system, across all visuals created by its AI products. Simultaneously, OpenAI released a free public verification tool that helps users determine if an image was generated by its platforms, specifically ChatGPT, the OpenAI API, or Codex.

The new verification tool operates by scanning for two distinct signals: SynthID and C2PA (a technical standard for image provenance). Users can upload images in PNG, JPG, or WEBP formats to check for these signals. OpenAI advises users to crop screenshots tightly and avoid uploading files containing multiple images to ensure the most accurate detection. The company indicated plans to expand this tool to support other AI generators in the future.

While C2PA provides readable metadata, it can be prone to manipulation. SynthID, developed by Google DeepMind, is engineered to persist through common modifications such as resizing, cropping, or screenshotting. By combining both methods, OpenAI aims to create a more resilient provenance framework. However, the efficacy of these measures remains limited to content created by tools adopting these standards, potentially leaving imagery from less secure models undetected.

OpenAI announced two major initiatives on May 19, 2026, to address the rise of AI-generated deepfakes and misinformation. The company partnered with Google to integrate SynthID, an invisible watermarking system, across all visuals created by its AI products. Simultaneously, OpenAI released a free public verification tool that helps users determine if an image was generated by its platforms, specifically ChatGPT, the OpenAI API, or Codex.

The new verification tool operates by scanning for two distinct signals: SynthID and C2PA (a technical standard for image provenance). Users can upload images in PNG, JPG, or WEBP formats to check for these signals. OpenAI advises users to crop screenshots tightly and avoid uploading files containing multiple images to ensure the most accurate detection. The company indicated plans to expand this tool to support other AI generators in the future.

While C2PA provides readable metadata, it can be prone to manipulation. SynthID, developed by Google DeepMind, is engineered to persist through common modifications such as resizing, cropping, or screenshotting. By combining both methods, OpenAI aims to create a more resilient provenance framework. However, the efficacy of these measures remains limited to content created by tools adopting these standards, potentially leaving imagery from less secure models undetected.