TEL AVIV – CyberWell, a trusted partner of Meta (Facebook, Instagram and Threads), TikTok and YouTube, and a leading organization in digital compliance solutions for hateful content, released new guidance for social media platforms on improving AI-led content moderation. The guidance forms part of CyberWell’s public comment submitted this month in response to Meta’s Oversight Board’s call for input on the use of large language models (LLMs) in content moderation. CyberWell outlines five recommendations to help Meta and the broader social media and tech industry improve the detection and removal of policy-violating content, including antisemitism and online hate, at scale while protecting freedom of expression.
“AI gives platforms an unprecedented opportunity to improve the enforcement of their own rules at scale, moving beyond slur lists toward context-based decisions that can help reduce human error. However, expert input is crucial to ensuring these systems can effectively address specific forms of online hate,” said Tal-Or Cohen Montemayor, CEO and Founder of CyberWell. “Antisemitism constantly evolves through coded language, symbols, cultural references and forms of ‘algospeak’ designed to evade automated detection. AI moderation systems must draw on contextual, linguistic and subject-matter expertise to recognize the antisemitic message, not just the words, while protecting legitimate expression. Social media platforms have also struggled to draw the line between anti-Jewish scapegoating, a classic form of antisemitism, and misinformation and disinformation campaigns that receive different moderation treatment.”
To address these challenges, CyberWell outlined five recommendations for Meta and the broader tech industry to improve LLM-based content moderation: (1) Work with subject-matter experts to improve model training and evaluation using real-world examples and annotated datasets; (2) Strengthen moderation across platform features and multimodal formats, including comments, usernames, images, videos and audio; (3) Improve multilingual accuracy through testing that accounts for dialects, slang, irony and coded language; (4) Establish clear escalation pathways for human review of complex or context-dependent content, particularly following violent incidents or during rapidly evolving global events; and (5) Increase transparency through regular reporting on LLM performance, enforcement outcomes, error rates and appeals, while providing users with clearer explanations of content removals.
CyberWell's expert submission highlighted real-world examples from its research illustrating the challenges that AI-based moderation systems must address when identifying antisemitic content, particularly when users employ coded language, cultural references, linguistic manipulation, and other tactics designed to evade detection.
One example showed an AI-generated video of Adolf Hitler dancing in front of Auschwitz, with a juice box emoji used as a coded reference to Jews. Other examples included an Arabic-language post shared by a news account that inserted dots between letters to evade automated moderation while amplifying Hamas messaging and praising a terrorist attacker. The submission also documented images and GIFs used in comment sections to mock the Holocaust and convey calls for violence against Jews.
“The future of effective Trust & Safety will depend on combining advanced technology with reliable data and real-world expertise,” Cohen Montemayor said. “Our working relationship with Meta and other major social media platforms demonstrates what that can look like in practice. The potential to scale effective moderation through AI makes it even more important to deepen relationships with trusted experts and partners who understand technology and digital policy. CyberWell is turning its expertise in how antisemitism and online hate evolve into practical datasets, evaluation tools and technological solutions that can scale, helping platforms optimize their moderation systems to identify policy violations, including nuanced and coded forms of hate, and enforce their rules more accurately and consistently.”
CyberWell is an independent, tech-based nonprofit combating the spread of antisemitism online, operating globally. Its AI-powered technologies monitor social media in English and Arabic for content promoting antisemitism, Holocaust denial and incitement to violence against Jews and their allies based on the International Holocaust Remembrance Alliance’s (IHRA) working definition of antisemitism. Its analysts review and report this content to platform moderators and index verified posts in the first-ever open database of antisemitic social media posts, cataloging it openly for transparency at app.cyberwell.org. Through partnerships, education and real-time alerts, CyberWell partners with social media platforms and digital service companies to help them enforce their policies more effectively, promoting proactive steps against online Jew-hate. For more information, visit: cyberwell.org
