Dr Aviv Barnoy, Assistant Professor in Media and Communication at ESHCC, and his team have won first place overall at #TikkunTech, the European Commission’s hackathon on combating antisemitism and fostering Jewish life. Their winning prototype, 4Sight, uses AI and human analysis to identify emerging forms of coded antisemitic language on fringe online platforms before they become established on larger social media platforms.
Online antisemitism does not always use explicit or easily recognisable language. New coded expressions can emerge that appear harmless to outsiders or automated moderation systems while carrying a different meaning within particular online communities. Examples include “271k”, which is used as shorthand for the false Holocaust-denial claim that only around 271,000 Jews were murdered during the Holocaust, and “Cohencidence”, a play on “Cohen” and “coincidence” used in conspiracy discourse to imply hidden Jewish involvement in events. Such expressions illustrate why fixed lists of known terms can struggle to keep pace with changing online language.
During the one-and-a-half-day hackathon sprint, Barnoy worked with Kevin Leiba, Sophie Taïeb and a fourth team member to develop 4Sight from an idea into a working proof of concept. Barnoy had been encouraged by the organisers to apply to participate and was subsequently selected for the event.
An early-warning system
4Sight is designed as an early-warning system. Rather than searching only for already-known antisemitic keywords, its large-language-model-based reader examines online posts for recurring expressions that may represent new or re-emerging coded language. The proof of concept used content from fringe platform 4chan /pol/ as its discovery environment and compared candidate expressions with data from X, looking for indications that language appearing in fringe communities was subsequently surfacing on a larger platform. The system combines frequency and timing information with contextual AI analysis, while keeping a human analyst in the loop to decide whether the evidence actually supports interpreting a term as antisemitic. In the hackathon prototype, 4Sight analysed 7,058 4chan posts and compared its findings with 13,422 posts from X.
Foresight
The broader ambition is to provide foresight rather than only retrospective moderation: identifying potentially important linguistic developments early enough for trust and safety, risk-mitigation and content-moderation teams to investigate them before they become widespread. The prototype does not itself remove or report content; it produces evidence for human assessment. This human-in-the-loop design was intentional, particularly because coded expressions can be ambiguous and AI-generated interpretations can be wrong.
4Sight was selected as the overall winner of TikkunTech. The two-day European Commission event in Brussels brought together more than 120 participants working at the intersection of technology, research, civil society and policy to develop practical responses to online antisemitism. European Commissioner for Internal Affairs and Migration Magnus Brunner participated in the prize-giving ceremony on 30 September.
As part of the winning prize, the team will receive funding to further develop 4Sight and an all-expenses-paid visit to TikTok’s Transparency and Accountability Center in Dublin, where they will meet with TikTok teams to discuss the project and possible further collaboration and development. The team intends to continue working together on the prototype.
- Researcher
