Tackle it Right in the Source Code

In February 2026, Ana Kasparian was called antisemitic on X. Her response: “the goyim are waking the f*** up.” She pinned a post —

“Israel is evil, genocidal and has destroyed our country”

— and refused to apologize. (Mediaite, via JPost) Eight months later, she took the fight to the man she holds responsible for the accusation machine itself. On October 14, Kasparian devoted a Young Turks segment to ADL CEO Jonathan Greenblatt’s campaign to reshape AI chatbots. 

She played a clip of Greenblatt boasting about, in his words, “excellent collaboration with Silicon Valley” — working with the AI labs to, again his words, “tackle it right in the source code.” (JPost)

Her case: 

Greenblatt is operating as a foreign operative against Americans’ constitutional rights, and the antisemitism framing is a hoax, a guise — cover for concealing Israeli war crimes. Greenblatt isn’t hiding the collaboration. 

In an October 12 interview with CTech,

he described the ADL’s work with AI labs as deliberately behind the scenes: “by the time you are talking to the press, it is already too late.” He said the ADL got OpenAI’s Sora to generate antisemitic videos, handed OpenAI the evidence, and the next version blocked the capability.

 He ranked six AI models 

— ChatGPT, Claude, Gemini, Llama, DeepSeek, Grok — with Grok “the most antisemitic by a wide margin,” because, he said, X is “a hellscape of raw, toxic antisemitism.” 

OpenAI, Anthropic, and Google’s Gemini team, he said, are “very attentive” to the ADL’s input. Others, he added — “you can figure out for yourself which ones” — were less so. 

He also volunteered the ADL’s address book: “very good ties” with the Department of Justice, the FBI, the Department of Education, and the State Department. 

Read that list again. 

The organization writing the rules for what chatbots may say about Israel is on warm terms with the agencies that investigate, prosecute, and deport. 

The ADL’s March 2025 report, “Generating Hate,” put numbers to the campaign: 86 test statements, 34,400 queries across GPT, Claude, Gemini, and Llama. All four models showed measurable bias, the report found; Llama was worst, GPT and Claude showed the most anti-Israel bias, and the models refused Israel-related questions more often than others. 

Source: Anti-Defamation League " Generating Hate "

To its credit, the report concedes its own limitations — including the risk of conflating criticism of Israel with antisemitism. That concession is doing a lot of work, because the entire enterprise turns on exactly that conflation. 

Here is what the record does not show.

 Exactly one AI model change is documented from ADL pressure — the Sora example — and it rests solely on Greenblatt’s telling. No AI company has publicly confirmed changing a model because of the ADL. And there is no verified case of a person harmed by an ADL-influenced chatbot output. 

The machinery of punishment is real

 — students swept up in “Catch and Revoke,” the March 2025 State Department cable that put AI screening on student visas, with some 6,000 visas revoked — but that is a different machine. Conflating the two would be the same sloppiness the ADL is accused of. The pattern around it is well documented. 

The IHRA definition’s own lead drafter, Kenneth Stern, says it was never intended as a campus speech code. A unanimous bipartisan U.S. Commission on Civil Rights report last month affirmed that antisemitism is a serious civil-rights problem — and warned against weaponizing the charge against protected political expression. 

Wikipedia’s editors deemed the ADL “generally unreliable” on Israel-Palestine back in June 2024. And this is not abstract: people lose jobs when the accusation gets deployed. This publication has documented one such case at length 

— part one, part two, and part three. 

Greenblatt said it like a boast: by the time you’re talking to the press, it’s already too late. He meant the labs would already have complied.

 Read it instead as a warning. 

A private advocacy group is writing the rules for what billions of people are allowed to hear from their machines — behind closed doors, before the public ever knows. That is not fighting hate. That is capturing the referee, and calling the objections bigotry.