AI-generated antisemitic posts are about twice as likely to justify or encourage violence as antisemitic content created by users, according to evidence presented to the Royal Commission on Antisemitism and Social Cohesion.
CyberWell founder and chief executive Tal-Or Cohen Montemayor told the commission that a 300-post sample of AI-generated antisemitic content examined by the organisation had received more than 30 million views across major social media platforms, with 65 per cent of those views on Instagram. She said the proportion of AI content justifying violence had risen from around 15 per cent to 33 per cent.
She also warned that children were being exposed to Holocaust denial disguised as popular cartoon content. In one example, a My Little Pony character was shown saying it was impossible to bake six million muffins, a coded reference used to deny the six million Jews killed in the Holocaust.
“You’re priming them [children] to also see the Holocaust and the story of the Holocaust through the lens of denial,” Cohen Montemayor told the commission.
The findings were contained in an expert report commissioned by the royal commission on the spread and development of antisemitism online.
Cohen Montemayor said social media companies had become better at identifying explicit Holocaust denial but continued to struggle with newer forms of antisemitism involving coded language, memes, conspiracy theories and mockery of Holocaust survivors and victims of hate-motivated violence.
She said automated moderation systems were often unable to recognise euphemisms and images designed to avoid detection.
“The challenge that all digital platforms, including social media companies face, is not one of comprehensive policy, it’s a challenge of scalable enforcement of their own rules,” she said. “As antisemitism evolves, platforms must evolve with it.”
She called on technology companies to create specialised datasets that separate different forms of antisemitism from other categories of hate speech. Without such data, she said, companies could not properly measure antisemitism, determine how recommendation systems spread it or train artificial intelligence to recognise it.
Cohen Montemayor also singled out comment sections as a major weakness in social media moderation.
“The Wild West of content moderation is the comments section,” she said.
She said antisemitic abuse frequently appeared in replies beneath posts that were neutral or did not breach platform rules, allowing hate to spread even when the original material remained compliant. Coded phrases, memes and euphemisms were increasingly being used to bypass automated systems while still conveying antisemitic messages, she said.
Cohen Montemayor urged platforms to strengthen moderation of comments, publish clearer information about enforcement and work with organisations that have linguistic, cultural and specialist knowledge of antisemitism.
She warned that the growing use of AI to moderate online material would produce weaker results unless expert knowledge was built into the systems.
“Building AI moderation systems meant to tackle nuanced and specific forms of hate speech and harassment on a global scale without the input of experts in emerging hate narratives is an oversight that will yield poorer results of an already exhausted and underfunded trust and safety ecosystem,” she said.
Cohen Montemayor, who is US-born and based in Israel, founded CyberWell in 2022. It is an independent non-profit organisation that monitors antisemitic material in English and Arabic, including Holocaust denial and incitement to violence. It reports verified posts to social media companies and maintains a public database tracking antisemitic content and platform responses.