Stay informed
AI-generated antisemitism is exploding online and social platforms are struggling to stop it: report
AI-generated antisemitic content is spreading quickly across the global social media platforms, drawing tens of millions of views while skirting moderation systems that struggle to keep up with the coded hate, according to a new report .
Between January 2025 and this February, analysts at CyberWell, an independent Israel-based non-profit whose mission is to combat online antisemitism, identified 307 AI-generated antisemitic posts in English on five major platforms — TikTok, YouTube, X and Meta’s Facebook and Instagram.
Those posts piled up more than 30 million views, over 2.8 million interactions (likes, shares, comments) across the platforms, most of it on video-based TikTok, YouTube and Instagram.
CyberWell found that more than 75 per cent of the posts fit into one of three narratives — “depictions of Jews as greedy or money-obsessed, Holocaust-related hate speech, and event-driven violent rhetoric against Jews.”
The organization also found a noticeable “sharp inflection point” in June 2025, which it attributed to Israel’s 12-day war on Iran. In the latest round of war on Iran carried out by Israel and the U.S., founder and CEO Tal-Or Cohen Montemayor told National Post in an interview that they’ve witnessed generative AI used to not only push antisemitic content but to promote “misinformation and disinformation.”
Alarmingly, more than a third of the content “glorifies, justifies, or calls for violence against Jews,” and those posts accounted for more than 33 per cent of the total views and 41 per cent of engagement by users.
Cohen Montemayor said she was very surprised to see that AI-generated content is twice as likely as user-generated content to be violently antisemitic.
“When you have generative AI content that’s literally fetishizing or justifying violence against Jews at a rate that’s twice as high as user-generated content … that’s very worrying to us,” Cohen Montemayor.
And for her, it raised a troubling, somewhat dystopian question about the algorithms that govern humanity’s digital spaces: “Is there something in the design or in the system that AI recognizes AI and is more likely to amplify that content in our future?”
Unlike traditional extremist content, much of the material identified in the report was designed to appear humorous, iconic or satirical — with many posts hashtagged as such — allowing it to spread before moderation systems could intervene.
“I think that the violence being the threshold by which any kind of tech platform is training their model to identify a problem doesn’t necessarily address the way that racism and hate show up in a social world,” Cohen Montemayor said.
The report found TikTok accounted for the largest share of the AI-generated antisemitic content in the dataset (35 per cent), while Instagram generated the majority of views (62 per cent) and engagement (92 per cent).
But as a result of existing explicit AI content moderation policies, those platforms (and Meta’s Facebook) also had the much higher rates of removal — 88 per cent and 67 per cent respectively — than YouTube (28 per cent) and X (20 per cent), platforms without similar policies.
Still, CyberWell reasoned that “high levels of engagement and views” the posts generated would suggest enforcement and removal occurred after they proliferated online. And while violent content is more likely to be removed, posts that used “implicit framing or coded language” to perpetuate the narratives stayed online longer.
Complicating removal is that existing content moderation classifiers — a “more primitive AI,” according to Cohen Montemayor — were trained on user-generated content and are more effective at identifying it than identifying content created by their superior AI kin.
“The infrastructure that the platforms have set up is simply exhausted by this proliferation of AI-generated hate content,” she explained.
Removal is important, the CEO said, because “the rules that govern these platforms matter.”
“It may occur to people that trust and safety or content moderation is in the weeds or its niche, but it’s as important as the laws that govern our streets,” she said.
The findings arise as antisemitism has seen a sharp increase in Canada, where Jewish advocacy organizations, law enforcement and political leaders have repeatedly warned that online rhetoric is increasingly spilling into the real world in the form of intimidation, harassment, threats and outright violence against Jewish people.
It also comes amid a growing debate within Canada about how to regulate online harms.
And while Ottawa has proposed legislation to address that and AI, Cohen Montemayor said any efforts to combat antisemitism focused only on educating people “without learning from what the tech and digital platforms are advertently or inadvertently pushing into their platforms is not a full stop solution.”
“Any strategies that are focused on confronting this problem and stemming it must also learn from the type of antisemitism that is being pushed by the tech platforms, including social media and the LLMs (Large Language Models) of today.”
In her view, to effectively govern the platforms and the AI, governments need a baseline framework that establishes “transparent guardrails” and a requirement to “fine-tune” platform trust and safety.
As for the platforms, CyberWell recommends that they explicitly apply hate speech and violence policies to AI-generated content, improve detection of the coded antisemitic narratives and remove harmful content before it can reach millions of people.
Our website is the place for the latest breaking news, exclusive scoops, longreads and provocative commentary. Please bookmark nationalpost.com and sign up for our newsletters here.
Unpublished Newswire
Comments
Be the first to comment