The five winning participants of the Safety Tech Challenge Fund were awarded an initial grant of £85,000 from the government to invest in developing solutions that would help to improve the detection and prevention of child sexual abuse material in end-to-end encrypted environments, whilst upholding user privacy.
They successfully produced their proof-of-concept tools by March 2022. We now have the chance to learn what progress has been made since the Challenge Fund ended.
Cyacomb
Founded in 2016, Cyacomb seeks to make the online world a safer place. They do this by helping law enforcement, cloud hosts and social media companies to find, block and remove illegal images and videos. Their team of experts develops ground-breaking technology that pinpoints harmful content at source.
Cyacomb works with the world’s most prestigious law enforcement agencies to eradicate harmful content and bring perpetrators to justice. The team collaborates with the UK government and international organisations, including DSIT and GCHQ, supporting their key value of collaboration alongside rigour, challenge and working for good.
With the help of the Safety Tech Challenge Fund, the team were able to demonstrate that it is possible to protect children and uphold user privacy. By matching against databases of known child sexual abuse material, they’ve shown that there are solutions which offer a high degree of user privacy and security whilst providing safety benefits for victims of child abuse.
With the financial investment and access to particular expertise provided by the Safety Tech Challenge Fund, they were able to produce technology solutions which are now being used in a number of pilots for operational use.
"Working on the Safety Tech Challenge Fund helped us to develop, validate and test an idea against a real need with stakeholders who were experts on the subject, and prove that it actually had the potential to make significant positive change for society."
T3K.ai
T3K.ai produces AI-enabled solutions to screen, analyse and detect harmful documents, images and videos. By leveraging artificial intelligence to identify and remove bad actors, the AI classifiers that form the foundation of their solutions can process multiple media formats and offer the highest detection rate on the market.
Solutions include a law enforcement analytics platform, contactless fingerprint detection and a content recognition engine. The latter uses a media classification API solution to detect illegal and harmful content shared or hosted online.
Working through the Safety Tech Challenge Fund, the team behind T3K.ai applied their knowledge and experience of screening video and photo data for illegal content to this project, and focused on scanning for child sexual abuse material.
The team used a very granular child sexual abuse material classifier - meaning there were nine subclasses based on ages of children and the severity of the content visible in the images. They also combined a facial age estimation with the classifier to give an additional level of accuracy.
Using the government funding, T3K.ai were able to explore their idea, set up a team and dedicate time to developing their proof of concept over five months, as opposed to the two or three years such projects can often take.
Since the Safety Tech Challenge Fund ended in March 2022, T3K.ai has found new ways to train classifiers to reduce the time, efforts and resources required to detect child sexual abuse material in end-to-end encrypted environments. Their next step is to focus their project on new areas such as parental controls.
“The Safety Tech Challenge Fund really made it clear that it isn’t that you can only have one or the other – safety or privacy – but that it’s possible to create solutions that combine both of these aspects and make it possible to create more safety whilst still maintaining privacy.”
Galaxkey
Galaxkey is an end-to-end encryption solution, meaning that information is encrypted at its source and can only be decrypted at its destination. Using Galaxkey’s three levels of identity-based encryption, users can protect all their data and communications, even outside of their network.
Now, with the support of the Safety Tech Challenge Fund, they’ve applied their expertise to the detection of unknown child sexual abuse material in end-to-end encrypted spaces, with the aim of finding a solution that offers safety for children without compromising user privacy.
The team used their own end-to-end encryption technology alongside two others, namely Yoti’s age estimation and an AI-based Image Analyser. The end result was Lock Chat – an end-to-end encrypted instant messaging platform where they would detect and prevent CSAM pre-encryption.
Through the Safety Tech Challenge Fund, the team proved that child sexual abuse material could be detected pre-encryption, and that safety and privacy are not mutually exclusive in these environments.
Once the Lock Chat platform is launched, the team hope the benefit will be felt almost immediately by target organisations, such as schools, as well as existing customers who have interaction with social workers.
“The Safety Tech Challenge Fund was really good because it provided us a platform to challenge our limits in technology where we could extend that to create a safety net for our children.”
SafeToNet
SafeToNet is a UK-founded safety tech company that provides AI-powered software designed to safeguard children from online risks such as bullying, sextortion, abuse and aggression. It works to prevent young people from putting themselves at potential risk by filtering harmful messages being sent on social networks and messaging apps.
The app analyses and filters messages in real time, detecting risks and linking these to behavioural patterns to help identify areas of concern. On detecting a potential risk, it also offers mental wellbeing support and tools to help young people navigate their digital world.
SafeToWatch is the latest innovation from the team, created with the support of the Safety Tech Challenge Fund. This real-time image and video threat detection software helps partners protect their platforms from harmful and illegal content including unknown child sexual abuse material.
By working through the Safety Tech Challenge Fund, the team were able to create a prototype design of SafeToWatch and at the conclusion of the project, had an app version to demonstrate how this would work in the long term.
Since then, SafeToNet has partnered with the Internet Watch Foundation to complete testing and produce a performance evaluation paper. They are also working with companies to provide trial usage, so they can test and pilot themselves.
The Challenge Fund allowed the SafeToNet team to connect with like-minded organisations who were all looking to develop similar solutions. It provided them with access to key expertise from the ICO and National Crime Agency, and funding to aid the development of their solution.
"The Safety Tech Challenge Fund shows that the government wants to offer a level of support [that] opens doors for people to meet with other like-minded organisations, have conversations and access opportunities."
DragonflAI
Founded in 2018 and based in Edinburgh, DragonflAI uses edge-based Machine Learning (ML) computation to prevent indecent and harmful content from being streamed or uploaded to the cloud. DragonflAI design, build and train algorithms small enough to run on mobile devices, detecting and analysing content on-device to prevent it from being shared online.
Working together with Yoti – a digital identity company – they combined their ML model with Yoti’s age estimation model to develop a product that would enable the detection of CSAM content in end-to-end encrypted environments.
The team developed a product that would identify unknown CSAM in an end-to-end encrypted environment within a reasonable timeframe. They wanted to achieve on-device moderation to prevent the upload of unknown CSAM material to centralised servers.
Since the challenge fund ended, DragonflAI have continued to develop their technology towards a more effective on-device application.
The Safety Tech Challenge Fund provided a unique opportunity to work with experts in different categories of ML, opening up discussions for potential future collaborations. Without the support and funding provided, it wouldn’t have been possible for DragonflAI to explore the feasibility of their project.
The challenge fund has given exposure to companies like DragonflAI, providing the vital funding needed to develop and test cutting-edge technologies, helping to raise awareness and drive the UK safety tech industry forward.
“I strongly believe that continuing our focus on moderating in the end-to-end encrypted space for both known and unknown harmful content is incredibly important, especially for our vision of a safer internet.”
Credits:
Created with an image by aksonsat - "Simple website HTML code with colorful tags in browser view on dark background. Software developer programming code. Screen of code for overlay background."