AI-Generated Antisemitism: Abuse Trends and Safety Gaps Across Social Media and AI Platforms

Key Findings:

• CyberWell analyzed 307 AI-generated antisemitic posts from Facebook, Instagram, TikTok, YouTube, and X, posted between January 2025 and February 2026. 98.4% appeared from June 2025 onward, showing a sharp mid-2025 increase.

• The dataset generated over 30 million views and more than 2.8 million interactions, making it CyberWell’s most-viewed dataset to date.

• AI-generated antisemitism was concentrated on visual and video platforms. TikTok, Instagram, and YouTube accounted for 79.2% of posts. TikTok had the largest share of posts (35.8%), while Instagram generated the most engagement (64.9%) despite representing only 24.8% of posts.

• The three most common narratives were:Jews as greedy or money-obsessed: 33.2%; Holocaust-related hate speech: 21.5%; Event-driven violent rhetoric against Jews: 21.2%

• Under the IHRA framework, 75.6% of posts involved stereotypical or conspiratorial claims about Jews, while 33.2% involved calls for, glorification of, or justification of violence against Jews. CyberWell notes that violent antisemitic content was more than twice as likely to appear in AI-generated antisemitic content as in user-generated antisemitic content.

• Platforms removed 63.5% of the posts overall, but enforcement varied sharply: TikTok: 88.2% removal; Meta: 67%; YouTube: 28.1%; X: 20%

• CyberWell argues that high eventual removal does not equal effective enforcement, because many posts remained online long enough to accumulate large audiences before being removed.

• Much of the content evaded detection through coded language, humor, Jewish music or symbols, fake news formats, “satire” disclaimers, Holocaust mockery, and youth-oriented styles such as Disney-Pixar-style trailers or gaming audio.

• CyberWell recommends that social media platforms explicitly apply hate, harassment, and violence policies to AI-generated content; improve detection across video, audio, image, and comment formats; limit algorithmic amplification; detect reuploads through watermarks and creator attribution; train moderators on coded antisemitism; and clarify that “satire” or “dark humor” labels do not exempt hateful content from enforcement.

• CyberWell recommends that AI companies define harmful content and protected categories more clearly, expand red-teaming for coded and implicit hate, and increase transparency around safety guardrails and risk mitigation.

• CyberWell recommends that policymakers require stronger transparency and audit mechanisms for AI systems, coordinate AI governance with platform regulation, address youth exposure to harmful AI-generated content, and standardize ways to identify AI-generated media across platforms.