Deepfakes and AI: The Fight Against a $40 Billion Fraud

A Growing Credibility Crisis

The proliferation of images generated by artificial intelligence has plunged the world into an unprecedented credibility crisis. The line between real photos and digital creations has become so blurred that it is increasingly difficult to distinguish the authenticity of images. This uncertainty has major repercussions across various sectors, including media, social networks, and even the judicial system, where the veracity of photographic evidence is crucial. According to the Deloitte Center for Financial Services, economic losses due to AI fraud could reach $40 billion in the United States by 2027, a significant jump from the estimated $12.3 billion for 2023.

Developing Standards: A Necessary Response

In the face of this growing threat, the need for international standards to identify and combat deepfakes and other AI scams is becoming urgent. Touradj Ebrahimi, a professor at the École polytechnique fédérale de Zurich, is leading these efforts. He is working closely with standardization bodies such as the International Electrotechnical Commission (IEC), the International Organization for Standardization (ISO), and the International Telecommunication Union (ITU). Their goal is to create standards that will help differentiate authentic content from digital fabrications.

The JPEG Trust standards, introduced to enhance the authenticity of images and data, are at the heart of this initiative. The first standard, unveiled last year, proposes incorporating metadata into JPEG files to serve as trust indicators. Two new parts of JPEG Trust are currently under development:

• JPEG Trust Part 2: This section introduces a catalog of trust profiles and reporting templates. These tools can be used directly or adapted for specific workflows and applications.
• JPEG Trust Part 3: This part focuses on the digital watermarking of media, an essential technique for ensuring content integrity.

Ebrahimi emphasizes that these standards aim to provide verification tools to end-users without validating or labeling images at the time of their creation.

A Fragmented Fight

Despite these efforts, the fight against deepfakes and AI fraud remains fragmented. Companies often develop their own solutions, leading to a diversity of approaches within the industry. Some companies are coming together to establish de facto standards, but the question of which standard will ultimately prevail remains open. Ebrahimi highlights the importance of this question, as a dominant standard could harmonize efforts to combat fraud.

Initiatives to counter fraud in the multimedia space truly gained momentum in 2018, when the JPEG committee of the IEC and ISO began addressing this growing issue of inauthentic content.

Other Standards in Development

Alongside JPEG Trust, other multimedia standards are being developed. Among them is the Origin Profile, which proposes a framework for documenting the origin of digital content. This profile includes guidelines for creating and maintaining profiles detailing the content creator and their creation process.

Another standard in development is the Vocabulary for Expressing Content Preferences for AI, which aims to standardize use cases involving machine-readable opt-out options related to data extraction and AI training.

Finally, the H.MMAUTH is a multimedia content authentication framework based on the digital signature of data streams. This system allows users to confirm the authenticity of content by its creators through a trusted third party.