Skip To Content
JEWISH. INDEPENDENT. NONPROFIT.
Culture

Grok said Hebrew translation was disabled on X — but it’s not

Why did the social media platform’s built-in AI bot hallucinate something so easy to disprove?

Despite what you may have heard, Hebrew translation still works on X. But allegations that the platform had disabled translation for Hebrew went viral after Grok, the AI chatbot built into the platform, said Hebrew was disabled because posts in the language were likely to encourage violence. As it turns out, the AI was hallucinating — the real question is why.

The rumor seems to have started because a Hebrew post advertising a pop group’s new single, “I, Butterfly,” was not working with the translation tools on the site. An account with the name “Red Pill Media” — though the bio for the account only says “America First,” and does not link to any media site — took a screenshot of an error message pop-up saying that Hebrew was not supported “for this translation.” They then shared the photo with a caption alleging that Hebrew translation was gone because “Jews were calling for genocide on this app without getting suspended.”

In the comments of this post, someone tagged Grok to ask why Hebrew wasn’t available. “Translation from Hebrew was disabled because it often amplified inflammatory or policy-violating content, like calls for violence, to a global audience via inaccurate or literal renditions,” the bot replied. “It’s about platform integrity amid documented spikes in Hebrew hate speech.”

Many people took this as an official confirmation from X that Hebrew translation had been turned off.

But while engineers, and the platform’s owner, Elon Musk, often tinker with the AI’s responses — for example, Musk made Grok more right-wing, and programmed it to flatter him — it largely consumes responses on X itself as its training material, which means that it is easy to mislead it. This is particularly the case on new, viral topics that its programmers have not had time to put up safeguards around.

In the comments on the original post, users speculated as to why the translation wasn’t working, quickly coming up with nefarious explanations. One user posited, or joked, that there was a Mein Kampf excerpt in the caption. Others guessed that it was an effort to “protect hate speech” in Hebrew so that English speakers can’t condemn it or use it to criticize Israel.

The original post that Grok could not translate contained no hate speech at all. It simply lists the song’s composers and the members of the band. (The translation issue may have stemmed from the fact that the song’s title was in English, and mixing characters from different alphabets confused the translation software.) But that didn’t stop false ideas about what it said from circulating. This is likely how Grok came to its conclusion — by consuming and regurgitating the conspiracy theories that users had themselves generated.

Chatbots and AIs are prone to hallucinations like this because of the way that they are trained; they tend to use human-generated input as their main source of information, which means that they are easily influenced by people’s own thoughts, incorrect beliefs and conspiracy theories. (This is also why they are prone to spouting neo-Nazi talking points without safeguards; there’s a lot of those floating around on the internet that the programs learned from.)

In fact, the error message in the screenshot saying that Hebrew was not available for that translation was not actually part of X; it was a pop-up from Apple Translation, the iPhone’s built-in translation tool, which was probably also confused by the mixed alphabets. And Grok has elsewhere confirmed that Hebrew can be translated on X, and that mixed alphabets cause a glitch. Still, theories continue to swirl that Grok may be refusing to translate Hebrew posts that include hate speech as part of an effort to reduce outcry against Israel.

But whether or not it’s good for the Jews, it’s still perfectly possible to translate plenty of racist statements in Hebrew, and any other language.

Republish This Story

Please read before republishing

We’re happy to make this story available to republish for free, unless it originated with JTA, Haaretz or another publication (as indicated on the article) and as long as you follow our guidelines.
You must comply with the following:

  • Credit the Forward
  • Retain our pixel
  • Preserve our canonical link in Google search
  • Add a noindex tag in Google search

See our full guidelines for more information, and this guide for detail about canonical URLs.

To republish, copy the HTML by clicking on the yellow button to the right; it includes our tracking pixel, all paragraph styles and hyperlinks, the author byline and credit to the Forward. It does not include images; to avoid copyright violations, you must add them manually, following our guidelines. Please email us at [email protected], subject line “republish,” with any questions or to let us know what stories you’re picking up.

We don't support Internet Explorer

Please use Chrome, Safari, Firefox, or Edge to view this site.