How global standards could stop $40B in AI image fraud

A deepening credibility crisis now surrounds digital imagery, with the line between authentic photographs and AI-generated fakes becoming nearly impossible to discern. This erosion of trust carries profound implications for news reporting, social media, and even legal evidence, while also fueling a surge in financial fraud. According to Deloitte’s Center for Financial Services, losses from generative AI-enabled fraud in the United States could skyrocket to $40 billion by 2027, up sharply from $12.3 billion in 2023.

In response, major international standards bodies are moving to equip users and businesses with the tools needed to separate reality from fabrication. At the recent AI for Good conference in Geneva, hosted under the UN’s umbrella, the International Electrotechnical Commission (IEC) and the International Organization for Standardization (ISO) announced new additions to their JPEG Trust standards. These efforts aim to provide a common framework for verifying the authenticity of images and data.

“You now don’t know if something is really fake or not,” said Touradj Ebrahimi, a professor at the Swiss Federal Institute of Technology who is leading work on combating AI fraud and deepfakes alongside the IEC, ISO, and the International Telecommunication Union (ITU). “You need to see metadata and information to find it, to put it in the right context.”

The JPEG Trust standard, first introduced last year, embeds trust indicators directly into JPEG files via metadata. Two new parts are now in progress. JPEG Trust Part 2 offers a catalog of trust profile snippets and reporting templates. These can serve as ready-made or customizable foundations for specific workflows, such as broadcasting, digital cameras, or AI-powered content generation. JPEG Trust Part 3 introduces media asset watermarking.

Ebrahimi stressed that JPEG Trust is designed for end-user verification, not for labeling content at the point of creation. “Fraudsters will not label their content as AI,” he explained. “If somebody wants to break the law, they’re not going to break the law and follow the other law that says that content needs to be labeled.” The standard, he added, “is not a standard that tells you whether to trust or not trust a video and image. It gives you means so that you can decide based on your content and your profile, and if you want to trust it.”

Context also plays a critical role. “Trust is very context-dependent,” Ebrahimi noted. “Some people might trust something because of their profile, because of their context. And even in the same context, they might trust later in another context.”

Until now, efforts to counter deepfakes and AI scams have been fragmented, with companies pushing proprietary solutions within their own ecosystems. “So a big question is which standard is going to become dominant?” Ebrahimi said. The IEC and ISO recognize the urgent need for widely accepted industry standards.

The push to guard against multimedia fraud began in earnest in 2018, when the JPEG committee first flagged concerns about inauthentic content. “They recognized there is a flaw in that JPEG files were being used to spread fake imagery,” Ebrahimi said. Today, with AI-generated imagery flooding the internet, the challenge is to extend these standards to help users verify what they see.

Beyond JPEG Trust, other standards are under development as part of the IEC and ISO’s AI and Multimedia Authenticity and Standards initiative. These include an originator profile framework for documenting the creator and creation process of digital content, a standardized vocabulary for machine-readable opt-outs related to text and data mining and AI training, and H. MMAUTH, a framework for authenticating multimedia content through digital signatures verified by a trusted third party.