Background: Child sexual abuse (CSA) investigations involving the seizure of electronic media (videos, images, and audio recordings) can require the painstaking and labour intensive forensic examination of thousands of files in order to locate information about victims and/or offenders. The often graphic nature of the content can also have serious adverse consequences for investigator mental health, and has accordingly led to the adoption of strategies that seek to limit exposure where possible. While the adoption of such strategies is necessary, they may also restrict the quantity and quality of intelligence that can be extracted and analysed from seized evidence.

Objectives: This paper introduces an automated software system designed to tackle this problem. This system uses state-of-the-art AI tools to (1) identify all persons (victims and offenders) contained within large troves of seized media files, and (2) establish the connections between offenders, their co-offenders (those who share and/or produce CSAM together) as well as with their victims.

Method: The software system, the Biometric Analyser and Network Extractor (BANE), uses a novel processing methodology to extract and match multiple biometric attributes (face and voice) from people contained within media files seized during multiple Australian CSA investigations, as well as their attendant metadata (geolocation, camera sensor data, and hash values).

Results: BANE match data are mapped using social network analysis to reveal the complex network of connections between offenders and their victims (within and across investigations). Additional modelling approaches are used to provide investigators with rapid, live intelligence about who to target (and in which files) to most effectively dismantle these networks.

Conclusions and key learnings: We discuss the implications of this work - offering insight into the ways this offending type is organised socially, as well as its practical implications for law enforcement. Future directions for this research are also discussed.