Insights
The potential of AI for time-critical CSAM investigations
By Shaunagh Downing, Fred Lichtenstein
Content warning | This blog post discusses child sexual abuse and exploitation, including child sexual abuse material (CSAM).
There is little doubt that perpetrators’ use of AI to target and exploit children online is overwhelming investigative resources. But could this emerging technology also become a vital part of investigators’ toolkits?
Here, we explore the very potential of AI for investigations into child sexual abuse material (CSAM) with Research and Development Engineer Dr. Shaunagh Downing and Research Development Lead Fred Lichtenstein.
AI is often talked about as a challenge for investigators, but is there a flip side? Does AI have the potential to aid investigations into CSAM?
Fred:
AI is already massively aiding investigations into child sexual abuse and exploitation.
Of course, we can’t ignore the fact that AI in the wrong hands poses a big threat to children. I know Shaunagh has addressed many of these risks before – like the misuse of Stable Diffusion to generate CSAM.
But AI tools can also help law enforcement tackle problems like these. For instance, they can help detect if an image has been generated or modified by AI – something that’s becoming harder for investigators to do by eye alone.
What are some other specific ways that law enforcement could use AI to move their investigations forward?
Shaunagh:
AI can analyse, organise, and summarise large amounts of data much more quickly than a human could. When it comes to prioritising investigative intelligence, these capabilities are especially helpful.
Investigators can also use AI to identify the most relevant evidence or intelligence to their cases, meaning that AI can make it easier to direct time and resources to the most high-risk, time-critical investigations.
Fred:
Exactly. Investigators often get huge volumes of data to analyse during CSAM cases, with not a lot of time to do it. It can be tough to know where to start or which intelligence to prioritise, based on which victims are in the most immediate danger.
AI models can help with this by creating a ‘triage pipeline’. By using classifiers for ‘unidentified victims’ and ‘high severity content’ to identify the most urgent cases, AI models can significantly speed up investigations and help investigators move forward with confidence.
AI can clearly support with simplifying and automating investigative tasks. Could it also help investigators get ahead of perpetrators?
Shaunagh:
When it comes to investigations, one of AI’s biggest strengths is that it can extract information which would otherwise be invisible to humans.
AI image generation is a prime example of this. If an image is really realistic, a human is unlikely to know for sure if it has been produced or modified with AI. Yet, it’s likely to have invisible signals that forensic AI tools can pick up on.
In a similar vein, AI can detect patterns in data that humans might not be able to see. This includes connections between devices and online accounts that can help investigators identify a perpetrator.
Fred:
Shaunagh and I actually saw an example of this in action at a conference earlier this year.
We watched an AI tool pick up on a clue in an image that was too blurry for a human to identify. This led to a potential location that the photo was taken in. It’s tools like these that can empower investigators to move quickly, and automate difficult, time-consuming parts of the job.
You might find interesting: AI image detection for investigators – a quick guide
What is CameraForensics doing to support investigators with AI?
Fred:
We’re working on lots of exciting projects which we’ll releasing in the coming months.
We're working on how to use AI more efficiently, with limitations in investigators’ computational power. To do this, we’re looking at how we can reduce the dimension of large data sets, without losing the data’s key principles.
These methods could ultimately help investigators process a large volume of images with AI quickly and accurately, so they can immediately find the most relevant connections for their cases.
Shaunagh:
We’ve also recently integrated Thorn Detect into some of our CameraForensics capabilities.
This is an AI-powered tool that allows investigators to find potential CSAM across devices – which is just another example of how AI is speeding up investigations into crimes against children.
It’s a really important tool that we’re pleased to connect investigators to.
Finally, how do you predict that AI will shape investigations into CSAM in the future?
Fred:
Over the next couple of years, I see agentic AI becoming a key tool for investigations. I think we’ll see an increase in tools that can carry out tasks for investigators and systems that they can interact back and forth with, as they would a chatbot or ChatGPT.
They might be able to ask AI to perform tasks like “Show me all the images taken in this location on this day”, for instance, or “Show me all the related images containing this logo”. There’s also huge opportunity for AI to help with investigator welfare, such as automatically classifying and describing harmful images.
Long-term, it’s hard to predict what might happen!
Shaunagh:
In the future, I see AI becoming a key collaborator in CSAM investigations. Not replacing human investigators, but enhancing their ability to act faster and more effectively.
These cases often involve overwhelming volumes of data, and AI will be able to help by automatically extracting important information, scaling analysis, enabling faster decisions and prioritising the most urgent risks. I also hope that AI will handle repetitive or traumatic tasks at scale, reducing the burden on investigators.
Read more on this topic: Investigators' wellbeing in CSAM cases – a Q&A with Rob Chitham
Dedicated to supporting investigators
If you’d like to read more about the latest threats to child victims online – from AI to the dark web – and the challenges these pose to investigators, download our Child Safety Online Report.
In it, you’ll also find the three steps that we at CameraForensics believe must be taken to help investigators create a safer world for children.
