Meta’s new AI detection tool isn’t fully working as advertised, according to a new report.
Meta debuted its first image generation model, Muse Image, earlier this week. As part of the debut, the tech giant also announced that all images produced by its models will include an invisible watermarking system called Content Seal. This signal remains intact even if the AI-generated image is “cropped, compressed, resized, or screenshotted” by the user, the company claimed. To help catch the content seal signal, Meta also announced that it is previewing an AI detection tool that will check if Muse Image generated the image.
However, in a report published on Friday, Reuters reporters found that the AI detection tool failed to identify more than half of the images produced after cropping. In an initial test, Reuters found that the tool was able to correctly identify all 40 images generated by Muse Image as being generated by AI. But when those images were cropped to half or one-third of their original size, the tool could only identify 55% of them as being generated by AI.
As generative AI tools become better at creating creepy deepfakes, detection becomes a difficult problem to solve. According to cybersecurity firm DeepStrike, the amount of AI-generated deepfakes online increased by about 900% annually from 2023 to 2025. However, detection capabilities have not fully advanced in tandem with this boom in popularity. Although commercial AI detection tools themselves are powered by AI, they are still plagued by mistakes, but previous research suggests that the average person’s ability to identify AI-generated content is comparable to a coin flip.
While true success may not be immediate, it is this gap that Meta aims to address with its new Content Seal and its detection tools.
Muse Image and its accompanying products were likely to be a big step forward for Meta, which has lagged behind its competitors in the AI space. Last year, Meta CEO Mark Zuckerberg decided there was no better time to play catch-up and announced a major AI turnaround effort. This turnaround plan included investing billions of dollars in research and development and poaching top talent from rivals across the industry, all in pursuit of the lofty goal of building better AI products and creating artificial superintelligence.
After putting billions more into AI and going through some more organizational restructuring, Meta announced the first big result of its labor in April: Muse Spark. The company announced that it plans to open source this unique model in the future, but it received mixed reviews. Another big event was this week’s debut of Muse Image.
But the debut of the image generation tool and its companion tools has been mired in controversy, and not just because of the Reuters report. Instagram users were alarmed to learn that an AI model could use photos from public profiles without explicitly seeking consent from the public profile owner. That functionality has now been removed.
The company is now turning its attention to its next big generative AI debut, a video generator called Muse Video. We hope the company can close the gaps in its detection tools and properly address users’ privacy concerns before this model is retired.
