AI tool detects child abuse images with 99% accuracy

One of the worst abuses imaginable is to use the internet to spread content that sexually exploits children. And in recent years, material about child abuse on the web has grown exponentially. In 2019, the National Center for Missing and Exploited Children in the US reported 69.1 million files — three times the 2017 levels and an increase of 15,000 percent over the previous 15 years. A new AI-powered tool is aimed at stemming the flow of abusive content, finding the victims, and identifying the offenders. In this article, we see more about this AI tool that will play a significant role in ending child abuse related crimes. Do you want to become an artificial intelligence developer? Find the best online AI certification course that suits your interest and start your journey today!

Knowledge of blog

  • Defining child sexual abuse material
  • How the AI tool functions
  • Final word


It’s essential to understand the urgent need to address this crime, and AI has stepped in to eliminate this inhuman crime. 

Defining child sexual abuse material

Child sexual abuse material refers to any content that describes sexually explicit activities involving a child. Visual depictions include photographs, videos, digital or computer-generated images. These images and videos are then circulated for personal consumption, involving the documentation of an actual crime scene. More recently, sexual abuse has begun to surface in live streaming. In such cases, individuals pay to watch a child’s live abuse through a video streaming service. Due to its real-time nature and the lack of digital evidence left behind after the crime, this type of abuse is incredibly difficult to detect.


How the AI tool functions

Image recognition isn’t a new task for the branch of artificial intelligence. The AI tool’s system searches for digital fingerprints, called hashes, to detect abusive content on a platform. When it finds these hashes, it compares them to a dataset of millions of abusive material images and videos. If the content has not been reported in advance, algorithms trained on abusive material will determine if it is likely to be child sexual abuse material. It is then queued up by moderation teams on the platform for review, and reported to the National Center for Missing and Exploited Children, which adds the content to its database. It’s not known precisely how this system detects types of abuse, to avoid tipping criminals off about possible ways to sidestep the technology. However, several types of machine learning are used within the system that’s been developed. The system also uses artificial intelligence to match faces and locations across the already vast, millions-image CAID database. This way, police officers can identify whether the image was taken from a young person or the geographic area, pulled from the GPS metadata image file, has already been linked to an instance of misuse. Recent related work has been using abusers’ hand markings and genetic patterns to identify them in images and videos. The data is trained in artificial intelligence to identify abuse from previous police investigations, where the images have been labeled with the type of offense committed. Only secure sites allow AI developers to access police data. A vital component of this is determining age. When someone approaches the age of 18, recognizing whether someone is an adult or a child is very difficult for a human being. That’s why the system can determine whether an image contains an adult or a child’s image through the previously labeled data. The detection system is set to have a confidence ratio of 99 percent. Once the system determines if an image can contain child abuse, it places all potential matches in groups based on their severity. Officers then receive thumbnails of the images that can be checked in batches, rather than individually. They must filter out images that are not in the right category. Although humans are needed to verify the decisions of the machine within the process – it’s not fully automated on purpose. The system allows investigators to get through multiple images quickly. Initial tests of the technology in a pilot scheme found that 200 images of potential abuse could be processed per minute by individual staff members, previously 18 images per minute. The result is that investigations can be carried out more quickly, with police estimates saying categorization is now possible within 30 minutes, which may have taken 24 hours previously. While AI is developing incredibly rapidly, it is still unable to provide the context needed to address more complex cases. Despite the early use of the tech in the real world, its potential is being looked out for other crime types and different still images. It can be used to spot knives or guns. It’s just about indecent images or children, it’s just about being able to take a much faster look at images and videos. It could be CCTV from a murder inquiry. Even that CCTV is now getting higher resolution and better quality. Still, in murder, you could have 30 different video systems that you downloaded.


Final Word

It is an ongoing challenge to identify and combat the spread of child sexual abuse material. Governments, law enforcement, NGOs, and industry all have a critically important role to play in protecting children from this horrific crime. While technology alone is not a panacea for this challenge to society, this work represents a significant step toward helping more organizations do this challenging work on a scale. Protect a child from the potential threat, enroll in cybersecurity training.