The use of the Internet to distribute CSEM is an abhorrent crime. Referrals from Online Service Providers are key to fighting CSE. OSPs, detection technologies and users reporting suspicious material are improving. However, this leads to an increase in the sheer volume of referrals coupled with the increase in the distribution of CSEM online that is pushing MS LEAs to their limits and affecting their their capacity to prevent harm to infants and children, rescue those in immediate danger, and investigate and prosecute perpetrators. The NCMEC process has improved LEA capability. But, a typical CSE case contains 1-3 TBs of video, 1–10 million images. Limited human resources, manual analysis and the 4,000% increase in referrals since 2014 obligates a new approach. GRACE will apply proven techniques in ML to the referral and analysis process while embracing the very technical, ethical and legal challenges unique to fighting CSE. GRACE will leverage resources already in place at EUROPOL and its 9 MS LEAs and attempt to provide results early, frequently and flexibly, prioritising easy wins in the research plan (e.g. deduplication). By applying Federated Learning approach to the challenge of optimising analysis and information flow, GRACE will enable cooperation between LEAs in improving their own capabilities and harness experiential knowledge. The results of GRACE will be handed back to EUROPOL and MS LEAs for unrestricted use in their missions, helping to ensure their future technological autonomy.

The increasing complexity of security challenges combined with the accumulation of significant amounts of digital data calls for better and more widespread use of Artificial Intelligence (AI) capabilities for law enforcement agencies (LEAs). AI can provide benefits to LEAs at all levels given the right understanding, tools, data and protection while increased awareness of criminal misuse is providing an immediate and concerning threat that must be tackled rapidly. Furthermore, a community that brings together LEAs, researchers, industry, security practitioners and other actors in the security ecosystem under a coordinated and strategic effort is essential for the realisation of these efforts into operational practices. STARLIGHT presents an inclusive and sustainable vision for increasing the awareness, capability, adoption and long-term impact of AI in Europe for LEAs. Five strategic goals underpin STARLIGHT’s approach: (1) Improve the widespread UNDERSTANDing of AI across LEAs to reinforce their investigative and cybersecurity operations and the need to uphold legal, ethical and societal values; (2) Provide opportunities to LEAs to EXPLOIT AI tools and solutions in their operational work that are trustworthy, transparent and human-centric; (3) Ensure that LEAs can PROTECT their own AI systems through privacy- and security-by-design approaches, better cybersecurity tools and knowledge; (4) Raise LEAs’ expertise and capacity to COMBAT the misuse of AI-supported crime and terrorism; and (5) BOOST AI for LEAs in Europe through high-quality datasets, an interoperable and standardised framework for long term sustainability of solutions, and the creation of an AI hub for LEAs that supports a strong AI security industry and enhances the EU’s strategic autonomy in AI. STARLIGHT will ensure European LEAs lead the way in AI innovation, autonomy and resilience, addressing the challenges of now and the future, prioritising the safety and security of Europe for all.