Is Meta Hallucinating Its Holocaust Denial Ban?

A good hate content policy on a social media platform means nothing if the platform does not have adequate enforcement mechanisms or the will to use them. This was made very clear to me this week, about 90 seconds after I read an article that quoted Ben Good, Meta’s director of content policy, who gave his company a pat on the back for prohibiting Holocaust denial on its applications during a talk at this year’s “Hack the Hate” conference. Here are the gems I found on Facebook in that time:

• “Too many people have realized that the Hollow Cost fairy tale was created by the Epsteins of the world who spend their free time r*ping children and sacrificing them to Moloch.”
• “We know ((((who)))). And they all kvetch about their 6 gorillion. Fuck their fake hollow cost!”
• “The Holocaust Hoax: the real holocaust was not of the Jews, but of the German people. Everything was the opposite of what they told us…”
  I’ll spare you further examples.

You may notice that in two of these three examples, the authors tried to use codewords: “Hollow cost” instead of Holocaust, “6 gorillion” instead of six million; presumably in an effort to avoid having their hateful post flagged by any automated systems. But the last example, which literally used the phrase “Holocaust hoax,” shows that the authors probably didn’t need to bother with codewords.

Meta claimed back in 2021 that its automated systems caught 97% of hate speech that violated its terms of service. Maybe they did, then – although company documents leaked by a whistleblower, also in 2021, put that figure at 3-5%. But in January of 2025, Joel Kaplan, Meta’s chief global affairs officer, announced that henceforth its automated systems would be used only for “illegal and high-severity violations, like terrorism, child sexual exploitation, drugs, fraud and scams.” Notably absent from that list is hate speech, including Holocaust denial.

It gets worse. In February 2026, Meta’s official oversight board announced that Meta’s new LLM-based moderation system was only being used to evaluate content that was manually flagged by others. Unlike what Ben Good is reported to have said at the Hack the Hate conference, Meta does not appear to be proactively scanning its system for antisemitic content like it used to.

Just this week, the Center on Extremism at ADL released a report showing that Meta failed to detect hundreds of accounts that promote antisemitism and terrorist propaganda, including over a hundred associated with Nick Fuentes’ Groyper movement, and it didn’t respond in a reasonable amount of time when staff flagged those accounts and posts to Meta.

Holocaust denial is still listed as a “Tier 1” category of hate speech that Meta will remove from its platform according to its Hateful Conduct policy. But if Meta will be cutting back on its human content moderators and has already chosen to not look proactively for hate speech with its automated systems, how much does that really matter for the fight against hate? X has already turned into a cesspool of hatred. I worry that Facebook and Instagram are already walking down that dark road as well.