The Home Office and Microsoft have launched a new technique that uses artificial intelligence to identify and block child grooming conversations online.
Named Project Artemis, its development began at a hackathon co-hosted by the department and the company in November 2018, and it will be licensed free of charge to small and medium sized technology companies to help them stamp out child grooming on their platforms.
The model automatically flags conversations that could be taking place between groomers and children and passes on details to the relevant law enforcement agency.
Home Secretary Priti Patel said: “Predators must get the message loud and clear, that there is no safe space to groom children for abuse.
“We are committed to stamping out this vile crime and this technique is just one part of that. Through collaboration with international partners and industry we are leading a worldwide effort to keep children safe from abuse.”
Conversation analysis
The prototype of the technique was developed in Seattle in 2018. Engineers from Microsoft, Facebook, Google, Snap and Twitter worked for two days analysing thousands of conversations to understand patterns used by predators.
Since then, engineers have worked through technical, legal and policy aspects, analysing thousands more instances of grooming conversations to develop the technique. The work was led by a cross industry group made up of Microsoft, The Meet Group, Roblox, Kik, Thorn and others.
This group was spearheaded by academic Dr Hany Farid who had previously worked to develop a tool which assisted in the detection, disruption and reporting of child exploitation images.
Licensing and adoption of the technique will be handled by Thorn, a charity that focuses on harnessing the power of technology to protect children online.
The Home Office is also working with the Joint Security and Resilience Centre (JSaRC) to develop tools to identify and block livestreamed child sexual abuse and pledged £300,000 in May to further develop capabilities.
Image from GOV.UK, Open Government Licence v3.0
