Combatting the spread of child sexual abuse content

Every five minutes an analyst at the Internet Watch Foundation finds and removes an image or video that qualifies as child sexual abuse material (CSAM).

As part of our voluntary work, Trilateral’s data science team participated at a Hackathon organised by the Internet Watch Foundation aimed to create “tech for good” innovative solutions to stop the distribution of child sexual abuse imagery online.

Offenders often use advanced techniques to disguise and circulate child sexual abuse content, therefore, it happens that manual reporting and removal often cannot keep up with the high volume of abuse content distributed online.

Organisations dedicated to eradicating these materials on the web need advanced technologies to help in quickly and efficiently identifying where the abusive content is posted. When the Internet Watch Foundation hosted a one-day Hackathon in London, we provided our data science team with an opportunity to volunteer and support the foundation in achieving their goal.


Natural language processing to combat the spread of child sexual abuse content

Our approach to the challenge was to use natural language processing to recognise child sexual abuse content in file names and descriptions of images and videos. We built algorithms in collaboration with another participant that would automatically classify whether a description was child sexual abuse content or not. We used both supervised and semi-supervised techniques to train the algorithms and also relied on state-of-the-art deep learning algorithms.

CSAM inside

In all our technological solutions, we assess ethical considerations to guarantee algorithm transparency. By identifying potential bias, we adjusted the training data set to minimise the effects of expected potential bias. Our work in algorithmic transparency has been instrumental in our technology development, for example such approach has led the CESIUM application development, which provides new risk assessment and identification tools to prevent and combat child exploitation.


All solutions and intellectual property developed at the Hackathon were donated to the Internet Watch Foundation to support their work. The machine learning tool created as part of the Hackathon challenge enables analysts to more efficiently remove child sexual abuse material.

Learn more about our research in the field of

Law Enforcement and Community Safeguarding

Let's discuss your career