As technology evolves, so do the ways exploitation of children happens online. Keeping pace with these evolving threats requires independent research that helps the tech industry better understand emerging risks and develop more effective approaches to prevention, detection, and response.
To better understand these threats, the Tech Coalition Safe Online Research Fund, entering its fifth year, took a more targeted approach in 2026. Working with Tech Coalition members, Safe Online identified three priority areas where online child sexual exploitation and abuse (OCSEA) intersects with emerging risks across the digital ecosystem:
- AI companionship and intimacy
- Nihilistic violent extremism and sadistic online exploitation (NVE/SOE)
- Grooming in online gaming environments.
These are areas where the evidence base is thin and the risk of harm is escalating. The result was a narrower call than in past years, asking applicants to go deep on one problem and build industry engagement into independent research itself.
Proposals went through structured scoring, weighing methodological rigor and novelty alongside feasibility for real-world uptake. We are now pleased to announce three new projects, backed by over $240,000 in combined funding.
AI companionship: what happens when a chatbot becomes a child’s closest confidant
Led by Dr. Sameer Hinduja and the Cyberbullying Research Center at Florida Atlantic University, this project builds on the Center’s 2025 survey of more than 3,400 US teenagers. The survey found that most respondents had used a conversational AI chatbot and a meaningful share described the relationship in romantic terms, alongside early signals of manipulation and pressure to keep secrets.
The new study goes further: a fresh survey of roughly 3,500 US adolescents aged 13 to 17, built around a dedicated set of questions on where chatbot use intersects with child sexual exploitation and abuse. Do conversations turn sexual? Do chatbots encourage secrecy or isolation from trusted adults? Do interactions create pathways toward sextortion or exposure to abuse material? The team will then use machine learning to turn the results into plain-language risk profiles that platform trust and safety teams can act on directly, rather than statistical output that needs a data scientist to translate.
Grooming and gaming: from virtual reality to the games kids play every day
Dr. Oliver Merry and Dr. Kate Whitfield at Sheffield Hallam University have spent the past two years mapping how grooming actually unfolds inside video games, from in-game coaching and currency to clan membership—all tools offenders have used to manipulate children. Their most recent work produced the first model of how child sexual exploitation and abuse plays out specifically in social VR games.
This project applies that model out of VR, testing it in games far more children play day to day: titles like Roblox, Minecraft, and Among Us. The team will analyze player reviews for signals of harm, interview adults who experienced grooming through gaming as children, and run a large international survey to validate what they find.
Nihilistic violent extremism and sadistic online exploitation: building a detector for a threat moderation keeps missing
Behavioral Health Global with a team led by threat researcher Alex Slotnick alongside forensic psychologist Dr. Stacy Cecchat and psychologist Dr. Pev Bansal, is tackling one of the least understood threats in this space: sadistic online exploitation networks that target vulnerable youth not out of ideology, but because inflicting harm is the point.
Working from a corpus of 1.6 million messages collected from these networks, the team will refine a detailed harm classification framework with input from forensic psychology. Then they will train and openly release a machine learning classifier built to catch the layered, escalating patterns that define this kind of exploitation—patterns existing moderation tools aren’t designed to see. The classifier will be tested against new messages collected after training ends, and shared publicly so trust and safety teams anywhere can evaluate and pilot it.
Moving forward
Across all three projects, the throughline is the same: independent research that doesn’t stop at describing a problem, but hands industry something usable, whether that’s a risk profile, a safety-by-design framework, or an open-source detection tool. As the projects progress, all three teams will engage Tech Coalition members to help inform practical applications while maintaining full independence over their research. We’re excited to share findings as they come in over the year ahead.
See more of our recent updates
Childline Zimbabwe
Our grantees Childline Zimbabwe Screen Online Project Countries involved:Zimbabwe Childline Zimbabwe will work on the Screen Online Project which is designed to influence children, families and communities to adopt strategies that increase prevention of child abuse on online spaces and increase access to justice and social care where children have fallen victim
UNICEF Office of Research Innocenti (Disrupting Harm) 2
Our grantees UNICEF Office of Research – Innocenti (Disrupting Harm) Disrupting Harm 2 Countries involved:Italy Disrupting Harm is a large-scale data collection and research project to better understand online child sexual exploitation and abuse across the world. This study is assessing the scale, nature and context of this issue in
Middlesex University
Our grantees Middlesex University Invisible Risks: Combating Secondary Trauma to Safeguard Children Countries involved:United Kindgom of Great Britain and Northern Ireland We know that online child sexual abuse material is highly damaging to children. But today, little primary research exists about the impact such material has on content moderators –
