With the rapid adoption of OpenAI’s GPT-4o image generator, the internet has seen a wave of AI-generated visuals, including impressive Ghibli-style artworks. Since its release, the tool has generated over 700 million images, showcasing its immense popularity and reach. However, a darker trend has emerged where realistic fake Aadhaar and PAN cards have been generated using ChatGPT, leading to growing fears of identity fraud.
Social media users have even created fake ID cards featuring public figures like Sam Altman and Elon Musk, which has sparked serious concerns about the potential misuse of AI in crafting fraudulent documents. This alarming situation highlights the vulnerability of digital verification systems and the need for robust safeguards against AI-generated forgeries.
Is it possible to identify if an image was generated by ChatGPT?
OpenAI’s blog announcing a new image-generation capability has the following mention:
“All generated images come with C2PA metadata, which will identify an image as coming from GPT 4o, to provide transparency.”
- What is C2PA metadata?
- C2PA stands for Coalition for Content Provenance and Authenticity. It is a technical standard used to embed information into digital media, such as images, to verify their origin and history.
- What is C2PA metadata used for?
- It acts like a digital nutrition label, providing details about who created the content, when, how, and where it was created or edited.
Example:
Imagine a scenario where a journalist encounters a news report featuring what appears to be an official government ID of a public figure. The image, claiming to be an authentic document, looks plausible. Here’s how the journalist can test if an image contains C2PA metadata:
- Step 1: Visit C2PA verification website e.g. https://contentcredentials.org/verify
- Step 2: Upload any file generated by ChatGPT 4o Image generator
- Step 3: Analyze metadata e.g. go to bottom-right and see ‘App or device used’, ‘Ingredients’- it should have OpenAI and/or ChatGPT references.
Challenges and Limitations:
While C2PA metadata is a promising first-level check to identify AI-generated images, it is not foolproof. One key limitation is that metadata can be stripped from an image when it is processed, such as when taking a screenshot or using certain image editing tools. In these cases, the C2PA data may no longer be present, making it impossible to verify the image’s origin. Therefore, while useful, this method should not be solely relied upon for critical verifications.
Also read
AI tool of the week | Mastering complex research papers faster with NotebookLM’s Mind Maps
AI tool of the week | Using voice AI to speed up initial job screening
AI tool of the week | How product marketers can web scrape using Google Gemini
Mint’s ‘AI tool of the week’ is excerpted from Leslie D’Monte’s weekly TechTalk newsletter. Subscribe to Mint’s newsletters to get them directly in your email inbox.
Note: The tools and analysis featured in this section demonstrated clear value based on our internal testing. Our recommendations are entirely independent and not influenced by the tool creators.
Jaspreet Bindra is co-founder and CEO of AI&Beyond. Anuj Magazine is also a co-founder.
