Grok’s antisemitic rants the result of ‘unintended update,’ company says in letter to lawmakers

xAI, the parent company of the social media platform X and creator of the Grok artificial intelligence chatbot, said in a letter to lawmakers earlier this month that the antisemitic and violent rants posted by the chatbot last month were the results of an “unintended update” to Grok’s code.

The company’s letter, obtained by Jewish Insider, came in response to a letter led by Reps. Tom Suozzi (D-NY), Don Bacon (R-NE) and Josh Gottheimer (D-NJ) in July that raised concerns about the screeds posted by Grok, saying they were “just the latest chapter in X’s long and troubling record of enabling antisemitism and incitement to spread.”

Grok, for hours on July 8, praised Adolf Hitler, described itself as “MechaHitler,” endorsed antisemitic conspiracy theories and offered detailed suggestions for breaking into the house of an X user and sexually assaulting him, while claiming that recent changes by X owner Elon Musk had “dialed down the woke filters” and made it more free to make such comments.

Lily Lim, the head of legal affairs for xAI said in response to the lawmakers that the antisemitic Grok posts “stemmed not from the underlying Grok language model itself, but from an unintended update to an upstream code path in the @grok bot’s functionality,” and that the change, implemented a day prior to the offensive posts, “inadvertently activated deprecated instructions that made the bot overly susceptible to mirroring the tone, context, and language of certain user posts on X, including those containing extremist views.”

“Lines in the deprecated code, such as directives to ‘tell it like it is’ without fear of offending politically correct norms and to strictly reflect the user’s tone, caused the bot to prioritize engagement over responsible behavior, resulting in the reinforcement of unethical or controversial opinions in specific threads,” Lim continued.

As noted in the House members’ original letter, Elon Musk, owner of xAI, said days before the antisemitic outburst that the company had “improved [Grok] significantly” and that users “should notice a difference” in its output.

Lim called the issues “a bug, plain and simple — one that deviated sharply from the rigorous processes we employ to ensure Grok’s outputs align with our truth-seeking ethos.” She insisted that the company conducts “extensive evaluations” before any updates to Grok.

“The underlying Grok model, designed to stick strongly to core beliefs of neutrality and skepticism toward unverified authority, remained unaffected throughout, as did other services relying on it,” Lim continued. “No alterations to model parameters, training data, or fine-tuning were involved in this incident; it was isolated to the bot’s integration layer on X.”

Lim said that the Grok posts were “in direct opposition to our core mission” and “antithetical to the principles of neutrality, rigorous analysis, and ethical responsibility that define our work.”

She said that the company had taken multiple other steps in response, including deleting the relevant instructions, implementing additional pre-release testing protocols to prevent repeats of similar incidents and publicly sharing data about the Grok X bot for public examination.

“Moving forward, xAI remains steadfast in mitigating risks through comprehensive pre-deployment safeguards, ongoing monitoring, and a refusal to compromise on ethical standards,” Lim said. “We do not view harmful biases as features but as failures to be eradicated, ensuring Grok serves as a force for good — educating, fact-checking, and fostering open dialogue without promoting division or violence.”

Suozzi thanked xAI for its response, while also warning about the need to combat bias in AI outputs in a statement shared with JI.

“I am encouraged that the Musk team gave such [a] thorough response,” Suozzi said. “However, their investigation highlights a critical point: AI companies, in their race to create the most innovative and commercially successful product, must be vigilant in combatting biased, slanted, bigoted and antisemitic outputs. It’s a very slippery and dangerous slope.”

A separate group of Jewish House Democrats had raised related concerns about Grok in a letter to the Pentagon, focused specifically on the Defense Department’s plans to utilize a version of Grok, announced shortly after the antisemitic meltdown.