Meta’s AI Detector Can’t Detect Images It Generated Itself, Report Finds

Meta’s new AI detection tool isn’t working entirely 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 generated by the model would include an invisible watermarking system called Content Seal. This signal would remain intact even when the AI-generated image gets “cropped, compressed, resized, or screenshotted” by users, the company claimed. To help with catching the Content Seal signal, Meta also announced that it was previewing an AI detection tool to check whether Muse Image generated an image.
But, in a report published on Friday, Reuters reporters found that the AI detection tool failed to identify more than half the images it generated once they had been cropped. In the first test, Reuters found that the tool correctly identified all 40 images generated by Muse Image as AI-generated, but once those images were cropped to half or one-third of their original size, the tool was only able to identify 55% as AI-generated.
As generative AI tools get better at producing uncanny deepfakes, detection becomes a trickier problem to solve. According to cybersecurity firm DeepStrike, the volume of AI-generated deepfakes online has experienced a roughly 900% annual growth from 2023 to 2025. But detection capabilities haven’t advanced completely in parallel to this boom in popularity. Commercial AI detection tools, themselves driven by AI, are still plagued with mistakes, while the average person’s ability to identify AI-generated content is no better than a coin toss, according to previous studies.
Though not a true immediate success, that’s the gap Meta is aiming to address with the new Content Seal and its detection tool.
Muse Image and its accompanying products were meant to be a major step forward for Meta, which has arguably been trailing its competitors in the AI space. Last year, Meta CEO Mark Zuckerberg decided there is no time like the present to try to catch up and announced a major AI turnaround effort. The catch-up plan included committing multibillion-dollar investments into research and development and poaching top talent from rivals all across the industry, all in pursuit of building better AI products and the lofty goal of creating artificial superintelligence.
An additional couple billion dedicated to AI and a few more restructurings later, Meta unveiled the first major fruit of that labor in April with Muse Spark, a proprietary model that it said it plans to open-source in the future and was met with a mixed reception. Another major one was this week’s Muse Image debut.
But the debut of the image generator and its accompanying tools has been mired in controversy, and not just because of the Reuters report. Instagram users were alarmed when they found out that the AI model could use photos from any public profile without explicitly asking the owner of said profile for their consent. That feature has now been removed.
The company is now eyeing its next big generative AI debut: a video generator called Muse Video. Here’s to hoping the company can bridge any gaps in detection tools and adequately address users’ privacy concerns before that model drops.

Facts Only

* Meta debuted its Muse Image model and announced an invisible watermarking system called Content Seal for all generated images.
* Content Seal is claimed to remain intact even when AI-generated images are cropped, compressed, resized, or screenshotted.
* Meta previewed an AI detection tool to check if Muse Image generated an image.
* In a Reuters report, the AI detection tool failed to identify more than half of the generated images once they were cropped.
* The tool correctly identified all 40 generated images as AI-generated initially.
* After cropping, the tool identified only 55% of the images as AI-generated in one test scenario.
* The volume of AI-generated deepfakes online experienced a roughly 900% annual growth from 2023 to 2025.
* Commercial AI detection tools are still plagued with mistakes.
* Instagram users were alarmed by the model's ability to use public profile photos without consent, leading to the removal of that feature.
* Meta is planning a next generative AI debut: Muse Video.

Executive Summary

Meta introduced its Muse Image model with an invisible watermarking system called Content Seal, designed to persist even when images are manipulated through cropping, compression, resizing, or screenshots. To facilitate detection of this signal, Meta previewed an AI detection tool for Muse Image. Testing of this tool revealed that it failed to identify more than half of the generated images once they were cropped. Specifically, in one test, the tool correctly identified all generated images as AI-generated initially, but only detected 55% as AI-generated after cropping them to half or one-third their original size. Furthermore, generative AI deepfakes are growing rapidly, with the volume experiencing roughly 900% annual growth from 2023 to 2025, while detection capabilities have not advanced at the same pace. This situation exists despite Meta's efforts with Content Seal and the detection tool, and concerns regarding user privacy arose when Instagram users discovered the AI model could utilize public profile photos without explicit consent, leading to the removal of that feature before the launch.

Full Take

The narrative surrounding the deployment of watermarking and detection tools highlights a fundamental tension between technological advancement, corporate control, and user autonomy. The inconsistency in the detection tool's performance—correctly identifying all images initially but failing significantly upon manipulation—suggests that the proclaimed mechanism (Content Seal) is either not robust enough against real-world adversarial attacks or that the detection logic itself suffers from systemic limitations when faced with compression artifacts. This points to a pattern where high-profile technological announcements prioritize narrative momentum over demonstrable, resilient security architecture. Furthermore, the preceding controversy regarding consent for image usage reveals a crucial dynamic: trust is easily eroded when systems are deployed without explicit user control, regardless of the technical safeguards put in place later. The pursuit of catching up in the AI space, marked by massive investment and talent acquisition, appears to be equally focused on creating proprietary generation capacity (Muse Image) as it is on mitigating risks associated with that capacity (detection tools). This dynamic suggests an underlying paradigm where competitive advantage is established through rapid deployment, forcing subsequent risk mitigation into a reactive posture rather than being built into the foundational design. What follows is the inevitable tension between speed of innovation and the necessity of reliable accountability in complex digital systems.
Patterns detected: ARC-0043 Motte-and-Bailey, ARC-0024 Ambiguity, ARC-0051 Systemic

Sentinel — Human

Confidence

The text reads like a typical, albeit slightly synthesized, news report that effectively weaves together product announcements, external reports, and corporate strategy into a coherent narrative about AI development hurdles.

Signals Detected

Moderate sentence length variance and organic flow.

Passionate framing regarding Meta's goals juxtaposed with specific data points; not overly sterile.

References to Reuters report, DeepStrike, and specific timelines suggest sourcing attempts, though attribution is general.

The core narrative flows logically from product launch to failure to detection strategy, typical of reporting structure.

Human Indicators

Use of contextual bridging between specific reports (Reuters) and broader corporate strategy (Zuckerberg's turnaround plan).

The inclusion of thematic shifts—from technical detection failure to privacy controversy to future product pipeline—demonstrates narrative weaving beyond simple data recitation.