Report: Instagram promotes antisemitic content
What changed after the moderation overhaul
The Nexus Project report links the rise in antisemitic material to a decision Meta made in January 2025. The company ended its work with third-party moderators and switched to a system in which users report violations themselves. Meta said the move was intended to allow more freedom of expression and reduce mistakes, arguing that the previous moderation system had gone too far and removed harmless posts.
The report’s authors say the changes made Meta’s approach to platform safety less proactive and put the responsibility for identifying and limiting hate speech, violent material and misinformation on users. Nexus Project says Meta often leaves up material even after users report it.
The authors argue that AI generation makes it faster to create posts that portray Jews in a malicious light. In their account, weaker moderation lets this material spread alongside conspiracy theories and other forms of hate. Instagram recommendations may lead users who show interest in such content to more antisemitic posts.
What new accounts found
To test how recommendations worked, Nexus Project created several new accounts and searched for material related to Jews. In one case, the top result for the search “synagogue” was an AI-generated video in which a rabbi welcomed the Antichrist. The video had more than a million views.
In less than an hour, almost the entire home page of a new account consisted of antisemitic posts. More than half the content came from verified accounts. The posts included claims that “Jews created LGBTQ” and a video urging goyim to “wake up.” Goyim is a word for non-Jews that is used in antisemitic statements.
Below antisemitic posts, Instagram suggested searches for conspiracy theories about “brainwashing goyim” and “dancing Israelis.” In its trending topics section, the platform recommended the Nazi marching song “Erika.” According to the report, it had been used in thousands of videos.
Viral posts often relied on edited clips from podcasts and other videos, making it seem as if Jews had said or supported things they had not actually said or supported. One example was the theory that the Jewish community secretly controls the world.
The report’s authors also point to ways people make money from this content. Accounts and creators promoted supplements, ebooks, courses and subscriptions. Meta, Nexus Project notes, may earn revenue from paid verification badges and ads placed alongside such material.
Meta’s response and journalists’ test
Meta said its efforts to combat antisemitism had not changed. The company said it had not removed specific safeguards and still prohibits harmful stereotypes about Jews, Holocaust denial, support for white supremacist organizations and dangerous organizations. It also said that using the word “Zionist” as a substitute for “Jews” or “Israelis” in hate speech is prohibited.
Meta called the report factually inaccurate and said it misrepresented the company’s rules. The company says the prevalence of hate-related violations has remained stable since 2025 and says it will continue taking action against posts that violate its rules.
Journalists who tested Nexus Project’s findings got similar results. Among the top results for “synagogue” was an AI-generated video showing a large figure resembling Satan, covered in blood, standing in a place of worship as Jewish men celebrated around it. A search for “Jew” returned numerous caricatures of Jews resembling Nazi propaganda; they were posted by accounts focused on conspiracy theories and hate. After “Jews” was entered, Instagram suggested searching for “jews controll everything”—with “control” misspelled, as shown in the original query.
A search for the song “Erika” returned numerous antisemitic posts. They included AI-generated videos of students and fans at a high school football game giving the Nazi salute to the song. Some accounts sharing it included links to download “Erika” ringtones in their bios.
Taken together, the videos found through these searches had millions of views. When journalists returned to the same account the next day, its feed was filled with even more hateful material.
Dataist's take: The tests described here point to a problem that goes beyond individual posts: after an initial interaction, recommendations may keep leading users to similar content. Meta disputes the report’s findings, so it is important to distinguish between the researchers’ claims, the journalists’ test results and the company’s position.
Daily AI news
Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.
Only what matters — every day
Follow on X