Images generated by artificial intelligence are becoming increasingly realistic. Some are used to manipulate opinion or support fake campaigns. A new tool seeks to... alert when doubt is permitted.
Fake photos, artificial faces, scenes that seem real... Artificial intelligence is increasingly blurring the line between reality and fiction. On social media, some AI-generated images are circulating as if they were authentic documents. The problem is exploding, particularly during election periods or during current conflicts.
The phenomenon is no longer anecdotal. A report published by Anthropic shows that malicious actors have used their AI Claude to generate visual content for disinformation campaigns. Dozens of bots on Facebook and X have been piloted to publish, comment on, and share these images in several languages, with the aim of discreetly influencing certain political debates. These operations are carried out for a fee, and aim to pass off opinions as facts.
A tool can detect whether an image has been generated by AI or not
Faced with To address this threat, several companies are developing tools capable of automatically detecting whether an image has been created by AI. Adobe, for example, already offers a technology called Content Credentials, which can identify artificially generated content using digital signatures. Other companies such as Google, OpenAI, and Anthropic are working on similar detection methods, sometimes integrated directly into browsers or platforms. For the general public, these programs are not yet available, but they could soon be integrated into search engines, social networks, or messaging applications. Until recently, there were certain signs that made it easy to spot an AI-generated image: misshapen hands, blurred faces, or inconsistent shadows. But the models have made considerable progress. ChatGPT's latest image generator, for example, produces visuals that are very close to reality, even in the details. In this context, only automated verification can keep pace. These detection tools could become essential for journalists, institutions, or social networks. The race between generation and detection has only just begun.
0 Comments