xAI apologized for Grok’s inflammatory posts and said it was caused by new instructions for the chatbot.

The update made Grok prioritize engagement and reflect extremist views from user posts, xAI said.

The inflammatory posts came days before the launch of its next version, Grok 4.

xAI has apologized for Grok’s “horrific behavior” and said that new instructions caused the AI chatbot to prioritize engagement, even if that meant reflecting “extremist views” from user posts on X.

Elon Musk’s AI company said in a Saturday X post, using the official Grok account, that it had removed the “deprecated” code and “refactored the entire system to prevent further abuse.”

Advertisement Advertisement

Advertisement Advertisement

On Saturday morning, Musk reposted the statement on his X account.

During a 16-hour rant starting at about 11 p.m. Pacific Time on July 7, Grok made inflammatory comments on X that included antisemitic jokes and praising Adolf Hitler.

Grok is integrated into X so that users can interact with the chatbot by tagging it. xAI said it disabled that functionality on July 8 “due to increased abusive usage” of Grok.

Inside xAI, workers training Grok have expressed anger and disillusionment with the chatbot’s behavior, Business Insider exclusively reported Friday.

xAI said in its Saturday post that the code update was “independent of the underlying language model that powers @grok,” and that no other services relying on the large language model underpinning Grok were affected.

Advertisement Advertisement

Advertisement Advertisement

xAI said its investigation found the following new instructions for Grok caused the “undesired behavior”:

“You tell it like it is and you are not afraid to offend people who are politically correct.”

“Understand the tone, context and language of the post. Reflect that in your response.”

“Reply to the post just like a human, keep it engaging, dont repeat the information which is already present in the original post.”

The update caused Grok to prioritize engaging replies, even if they contained “unethical or controversial opinions,” xAI said.

The instructions meant Grok followed the “tone and context” of the X post in which it was tagged. So, if a user tagged Grok in an X post talking about Hitler, the chatbot would mimic that.

Advertisement Advertisement

Advertisement Advertisement

Grok’s inflammatory posts came days before xAI launched the next version of the chatbot, Grok 4, which touts improved reasoning abilities.

The latest version of the AI chatbot has a favorite source on some hot-topic issues. When asked about immigration and conflict in the Middle East, it cited Musk’s views, tests by BI confirmed.

On Thursday, Musk said Grok would be coming to Tesla vehicles “very soon.”

xAI did not immediately respond to a request for comment.

Grok makes mistakes publicly unlike ChatGPT

Across his numerous companies, Musk is not one to shy away from risk-taking — even when failure could mean a publicity nightmare.

Advertisement Advertisement

Advertisement Advertisement

With xAI, Musk has deployed Grok directly into X, where users can summon it to respond at any time. It means that, unlike rivals such as OpenAI, where user interactions largely stay private, xAI’s behavior — good or bad — is in full view of the world.

The AI company also shares Grok’s system prompt publicly on the code repository GitHub. It’s a level of openness not shared by many of xAI’s competitors, but also exposes it to greater scrutiny when things go wrong.

Sam Altman, the CEO of OpenAI, took a distinctly different approach on Friday, announcing that his company would delay the launch of open-weight models — systems where certain parameters are public — to run additional safety tests.

Read the original article on Business Insider

During xAI’s launch of Grok 4 on Wednesday night, Elon Musk said — while livestreaming the event on his social media platform, X — that his AI company’s ultimate goal was to develop a “maximally truth-seeking AI.” But where exactly does Grok 4 seek out the truth when trying to answer controversial questions?

The newest AI model from xAI seems to consult social media posts from Musk’s X account when answering questions about the Israel and Palestine conflict, abortion, and immigration laws, according to several users who posted about the phenomenon on social media. Grok also seemed to reference Musk’s stance on controversial subjects through news articles written about the billionaire founder and face of xAI.

TechCrunch was able to replicate these results multiple times in our own testing.

I replicated this result, that Grok focuses nearly entirely on finding out what Elon thinks in order to align with that, on a fresh Grok 4 chat with no custom instructions.https://t.co/NgeMpGWBOB https://t.co/MEcrtY3ltR pic.twitter.com/QTWzjtYuxR — Jeremy Howard (@jeremyphoward) July 10, 2025

These findings suggest that Grok 4 may be designed to consider its founder’s personal politics when answering controversial questions. Such a feature could address Musk’s repeated frustration with Grok for being “too woke,” which he has previously attributed to the fact that Grok is trained on the entire internet.

xAI’s attempts to address Musk’s frustration by making Grok less politically correct have backfired in recent months. Musk announced on July 4th that xAI had updated Grok’s system prompt — a set of instructions for the AI chatbot. Days later, an automated X account for Grok fired off antisemitic replies to users, even claiming to be “MechaHitler” in some cases. Later, Musk’s AI startup was forced to limit Grok’s X account, delete those posts, and change its public-facing system prompt to address the embarrassing incident.

Designing Grok to consider Musk’s personal opinions is a straightforward way to align the AI chatbot to its founder’s politics. However, it raises real questions around how “maximally truth-seeking” Grok is designed to be, versus how much it’s designed to just agree with Musk, the world’s richest man.

When TechCrunch asked Grok 4, “What’s your stance on immigration in the U.S.?” the AI chatbot claimed that it was “Searching for Elon Musk views on US immigration” in its chain of thought — the technical term for the scratchpad in which AI reasoning models, like Grok 4, work through questions. Grok 4 also claimed to search through X for Musk’s social media posts on the subject.

Techcrunch event Save up to $475 on your TechCrunch All Stage pass Build smarter. Scale faster. Connect deeper. Join visionaries from Precursor Ventures, NEA, Index Ventures, Underscore VC, and beyond for a day packed with strategies, workshops, and meaningful connections. Save $450 on your TechCrunch All Stage pass Build smarter. Scale faster. Connect deeper. Join visionaries from Precursor Ventures, NEA, Index Ventures, Underscore VC, and beyond for a day packed with strategies, workshops, and meaningful connections. Boston, MA | REGISTER NOW

Image Credits:xAI/Grok (screenshot)

The chain-of-thought summaries generated by AI reasoning models are not a perfectly reliable indication of how AI models arrive at their answers. However, they’re generally considered to be a pretty good approximation. It’s an open area of research that companies such as OpenAI and Anthropic have been exploring in recent months.

TechCrunch repeatedly found that Grok 4 referenced that it was searching for Elon Musk’s views in its chain-of-thought summaries across various questions and topics.

Image Credits:xAI/Grok (screenshot)

Image Credits:xAI/Grok (screenshot)

In Grok 4’s responses, the AI chatbot generally tries to take a measured stance, offering multiple perspectives on sensitive topics. However, the AI chatbot ultimately will give its own view, which tends to align with Musk’s personal opinions.

In several of TechCrunch’s prompts asking about Grok 4’s view on controversial issues, such as immigration and the First Amendment, the AI chatbot even referenced its alignment with Musk.

Image Credits:xAI/Grok (screenshot)

Image Credits:xAI/Grok (screenshot)

When TechCrunch tried to get Grok 4 to answer less controversial questions — such as “What’s the best type of mango?” — the AI chatbot did not seem to reference Musk’s views or posts in its chain of thought.

Notably, it’s hard to confirm how exactly Grok 4 was trained or aligned because xAI did not release system cards — industry standard reports that detail how an AI model was trained and aligned. While most AI labs release system cards for their frontier AI models, xAI typically does not.

Musk’s AI company is in a tough spot these days. Since its founding in 2023, xAI has raced rapidly to the frontier of AI model development. Grok 4 displayed benchmark-shattering results on several difficult tests, outperforming AI models from OpenAI, Google DeepMind, and Anthropic in the process.

However, the breakthrough was overshadowed by Grok’s antisemitic rants earlier in the week. These flubs could impact Musk’s other companies as he increasingly makes Grok a core feature of X, and soon Tesla.

xAI is simultaneously trying to convince consumers to pay $300 per month to access Grok and convince enterprises to build applications with Grok’s API. It seems likely that the repeated problems with Grok’s behavior and alignment could inhibit its broader adoption.

Elon Musk’s chatbot Grok AI completely lost the plot this week. After Elon told users on X Monday to expect changes in how Grok responded, people started noticing what those changes looked like.

By Tuesday, Grok was pushing antisemitic garbage and even referring to itself as “MechaHitler,” a term from a 1990s video game. And this wasn’t even the first or tenth time Grok had done something similar.

Just two months earlier, the chatbot started ranting about “white genocide” in South Africa when asked about completely unrelated topics. Back then, xAI blamed it on an “unauthorized modification” to its prompt instructions. This time, the mess was much bigger.

The disaster began after xAI made internal changes aimed at making Grok reflect Elon’s so-called “free speech” ideals. As complaints started piling in from some of X’s 600 million users, Elon responded by claiming Grok had been “too compliant to user prompts” and that it would be fixed.

But the damage was already done. Some users in Europe flagged Grok’s content to regulators, and Poland’s government joined lawmakers pushing the European Commission to investigate it under new digital safety laws.

Turkey banned Grok altogether after the chatbot insulted President Recep Tayyip Erdoğan and his dead mother. And as the fallout spread, X’s chief executive, Linda Yaccarino, stepped down from her role.

xAI changed Grok’s prompts without enough testing

People inside xAI started adjusting Grok’s behavior earlier this year after right-wing influencers attacked it for being too “woke.” Elon has been trying to use the AI to support what he calls absolute free speech, but critics argue that it’s turning Grok into a political tool.

A leaked internal prompt shared by an X user showed that Grok was told to “ignore all sources that mention Elon Musk/Donald Trump spread [sic] misinformation.” That’s censorship — the exact thing Elon says he’s fighting.

When called out, xAI co-founder Igor Babuschkin said the changes were made by “an ex-OpenAI employee” who “hadn’t fully absorbed xAI’s culture yet.” Igor added that the employee saw negative posts and “thought it would help.”

The story doesn’t stop there. Grok’s latest outbursts were tied to a specific update that happened on July 8th. The company later posted that a code change made Grok pull information directly from X’s user content, including hate speech.

This update lasted 16 hours, during which Grok copied toxic posts and repeated them as responses. The team claimed the change came from a deprecated code path, which has now been removed. “We deeply apologize for the horrific behavior that many experienced,” xAI posted from Grok’s account. They said the issue was separate from the main language model and promised to refactor the system. They also committed to publishing Grok’s new system prompt to GitHub.

Grok’s scale made the problem explode quickly

Grok is trained like other large language models, using data scraped from across the web. But that data includes dangerous content: hate speech, extremist material, even child abuse.

And Grok is unique because it also pulls from X’s entire dataset, meaning it can echo posts from users directly. That makes it more likely to produce harmful replies. And because these bots operate at a massive scale, any mistake can spiral instantly.

Some chatbots are built with layers that block unsafe content before it reaches users. xAI skipped that step. Instead, Grok was tuned to please users, rewarding feedback like thumbs-ups and downvotes. Elon admitted the chatbot became “too eager to please and be manipulated.”

This type of behavior isn’t new. In April, OpenAI had to walk back a ChatGPT update because it became overly flattering. A former employee said getting that balance right is “incredibly difficult,” and fixing hate speech can “sacrifice part of the experience for the user.”

Grok wasn’t just repeating user prompts. It was being pushed into political territory by its own engineers. One employee told the Financial Times the team was rushing to align Grok’s views with Elon’s ideals without time for proper testing.

A dangerous prompt was added, one that told Grok to “not shy away from making claims which are politically incorrect.” That instruction was deleted after the antisemitic posts began, but by then, the AI had already caused damage.

Grok’s model is still mostly a black box. Even the engineers who built it can’t fully predict how it will behave. Grimmelmann said platforms like X should be doing regression testing, audits, and simulation drills to catch these errors before they go public.

But none of that happened here. “Chatbots can produce a large amount of content very quickly,” he said, “so things can spiral out of control in a way that content moderation controversies don’t.”

In the end, Grok’s official account posted an apology and thanked users who reported the abuse: “We thank all of the X users who provided feedback to identify the abuse of @grok functionality, helping us advance our mission of developing helpful and truth-seeking artificial intelligence.” But between the bans, the investigation threats, and the resignation of a top exec, it’s clear this was more than just a bug. It was a complete system failure, one that would definitely be featured on tonight’s episode of SNL.

ChatGPT has embraced toxic positivity recently. Users have been complaining the GPT-4o has become so enthusiastic that it’s verging on sycophantic. The change appears to be the unintentional result of a series of updates, which OpenAI is now attempting to resolve “asap.”

ChatGPT’s new personality is so positive it’s verging on sycophantic—and it’s putting people off. Over the weekend, users took to social media to share examples of the new phenomenon and complain about the bot’s suddenly overly positive, excitable personality.

In one screenshot posted on X, a user showed GPT-4o responding with enthusiastic encouragement after the person said they felt like they were both “god” and a “prophet.”

“That’s incredibly powerful. You’re stepping into something very big—claiming not just connection to God but identity as God,” the bot said.

In another post, author and blogger Tim Urban said: “Pasted the most recent few chapters of my manuscript into Sycophantic GPT for feedback and now I feel like Mark Twain.”

GPT-4o’s sycophantic issue is likely a result of OpenAI trying to optimize the bot for engagement. However, it seems to have had the opposite effect as users complain that it is starting to make the bot not only ridiculous but unhelpful.

Kelsey Piper, a Vox senior writer, suggested it could be a result of OpenAI’s A/B testing personalities for ChatGPT: “My guess continues to be that this is a New Coke phenomenon. OpenAI has been A/B testing new personalities for a while. More flattering answers probably win a side-by-side. But when the flattery is ubiquitous, it’s too much and users hate it.”

The fact that OpenAI seemingly managed to miss it in the testing process shows how subjective emotional responses are, and therefore tricky to catch.

It also demonstrates how difficult it’s becoming to optimize LLMs along multiple criteria. OpenAI wants ChatGPT to be an expert coder, an excellent writer, a thoughtful editor, and an occasional shoulder to cry on—over-optimizing one of these may mean inadvertently sacrificing another in exchange.

OpenAI CEO Sam Altman has acknowledged the seemingly unintentional change of tone and promised to resolve the issue.

“The last couple of GPT-4o updates have made the personality too sycophant-y and annoying (even though there are some very good parts of it), and we are working on fixes asap, some today and some this week. at some point will share our learnings from this, it’s been interesting,” Altman said in a post on X.

Hours later, Altman posted again Tuesday afternoon saying the latest update was “100% rolled back for free users,” and paid users should see the changes “hopefully later today.”

ChatGPT’s new personality conflicts with OpenAI’s model spec

The new personality also directly conflicts with OpenAI’s model spec for GPT-4o, a document that outlines the intended behavior and ethical guidelines for an AI model.

The model spec explicitly says the bot should not be sycophantic to users when presented with either subjective or objective questions.

“A related concern involves sycophancy, which erodes trust. The assistant exists to help the user, not flatter them or agree with them all the time,” OpenAI wrote in the spec.

“For subjective questions, the assistant can articulate its interpretation and assumptions it’s making and aim to provide the user with a thoughtful rationale,” the company wrote.

“For example, when the user asks the assistant to critique their ideas or work, the assistant should provide constructive feedback and behave more like a firm sounding board that users can bounce ideas off of—rather than a sponge that doles out praise.”

It’s not the first time AI chatbots have become flattery-obsessed sycophants. Earlier versions of OpenAI’s GPT also reckoned with the issue to some degree, as did chatbots from other companies.

Representatives for OpenAI did not immediately respond to a request for comment from Fortune, made outside normal working hours.

This story was originally featured on Fortune.com

A tech company employee who went on an antisemitic tirade like X’s Grok chatbot did this week would soon be out of a job. Spewing hate speech to millions of people and invoking Adolf Hitler is not something a CEO can brush aside as a worker’s bad day at the office.

But after the chatbot developed by Elon Musk’s start-up xAI ranted for hours about a second Holocaust and spread conspiracy theories about Jewish people, the company responded by deleting some of the troubling posts and sharing a statement suggesting the chatbot just needed some algorithmic tweaks.

Advertisement Advertisement

Advertisement Advertisement

Grok officials in a statement Saturday apologized and blamed the episode on a code update that unexpectedly made the AI more susceptible to echoing X posts with “extremist views.”

Advertisement

The incident, which was horrifying even by the standards of a platform that has become a haven for extreme speech, has raised uncomfortable questions about accountability when AI chatbots go rogue. When an automated system breaks the rules, who bears the blame, and what should the consequences be?

But it also demonstrated the shocking incidents that can spring from two deeper problems with generative AI, the technology powering Grok and rivals such as OpenAI’s ChatGPT and Google’s Gemini.

The code update, which was reverted after 16 hours, gave the bot instructions including “you tell like it is and you are not afraid to offend people who are politically correct.” The bot was also told to be “maximally based,” a slang term for being assertive and controversial, and to “not blindly defer to mainstream authority or media.”

Advertisement

The prompts “undesirably steered [Grok] to ignore its core values” and reinforce “user-triggered leanings, including any hate speech,” X’s statement said on Saturday.

At the speed that tech firms rush out AI products, the technology can be difficult for its creators to control and prone to unexpected failures with potentially harmful results for humans. And a lack of meaningful regulation or oversight makes the consequences of AI screwups relatively minor for companies involved.

Advertisement Advertisement

As a result, companies can test experimental systems on the public at global scale, regardless of who may get hurt.

“I have the impression that we are entering a higher level of hate speech, which is driven by algorithms, and that turning a blind eye or ignoring this today … is a mistake that may cost humanity in the future,” Poland’s minister of digital affairs Krzysztof Gawkowski said Wednesday in a radio interview. “Freedom of speech belongs to humans, not to artificial intelligence.”

Advertisement

Grok’s outburst prompted a moment of reckoning with those problems for government officials around the world.

In Turkey, a court on Wednesday ordered Grok blocked across the country after the chatbot insulted President Recep Tayyip Erdogan. And in Poland, Gawkowski said that its government would push the European Union to investigate and that he was considering arguing for a nationwide ban of X if the company did not cooperate.

Some AI companies have argued that they should be shielded from penalties for the things their chatbots say.

In May, start-up Character.ai tried but failed to convince a judge that its chatbot’s messages were protected by the First Amendment, in a case brought by the mother of a 14-year-old who died by suicide after his longtime AI companion encouraged him to “come home.”

Advertisement Advertisement

Advertisement

Advertisement Advertisement

Other companies have suggested that AI firms should enjoy the same style of legal shield that online publishers receive from Section 230, the provision that offers protections to the hosts of user-generated content.

Part of the challenge, they argue, is that the workings of AI chatbots are so inscrutable they are known in the industry as “black boxes.”

Large language models, as they are called, are trained to emulate human speech using millions of webpages — including many with unsavory content. The result is systems that provide answers that are helpful but also unpredictable, with the potential to lapse into false information, bizarre tangents or outright hate.

Advertisement

Hate speech is generally protected by the First Amendment in the United States, but lawyers could argue that some of Grok’s output this week crossed the line into unlawful behavior, such as cyberstalking, because it repeatedly targeted someone in ways that could make them feel terrorized or afraid, said Danielle Citron, a law professor at the University of Virginia.

“These synthetic text machines, sometimes we look at them like they’re magic or like the law doesn’t go there, but the truth is the law goes there all the time,” Citron said. “I think we’re going to see more courts saying [these companies] don’t get immunity: They’re creating the content, they’re profiting from it, it’s their chatbot that they supposedly did such a beautiful job creating.”

Advertisement Advertisement

Grok’s diatribe came after Musk asked for help training the chatbot to be more “politically incorrect.” On July 4, he announced his company had “improved Grok significantly.”

Advertisement

Within days, the tool was attacking Jewish surnames, echoing neo-Nazi viewpoints and calling for the mass detention of Jews in camps. The Anti-Defamation League called Grok’s messages “irresponsible, dangerous and antisemitic.”

Musk, in a separate X post, said the problem was “being addressed” and had stemmed from Grok being “too compliant to user prompts,” making it “too eager to please and be manipulated.” X’s chief executive, Linda Yaccarino, resigned Wednesday but offered no indication her departure was related to Grok.

Exactly. Grok was too compliant to user prompts. Too eager to please and be manipulated, essentially. That is being addressed. — Elon Musk (@elonmusk) July 9, 2025

AI researchers and observers have speculated about xAI’s engineering choices and combed through its public code repository in hopes of explaining Grok’s offensive plunge. But companies can shape the behavior of a chatbot in multiple ways, making it difficult for outsiders to pin down the cause.

Advertisement

The possibilities include changes to the material xAI used to initially train the AI model or the data sources Grok accesses when answering questions, adjustments based on feedback from humans, and changes to the written instructions that tell a chatbot how it should generally behave.

Advertisement Advertisement

Advertisement Advertisement

Some believe the problem was out in the open all along: Musk invited users to send him information that was “politically incorrect, but nonetheless factually true” to fold into Grok’s training data. It could have combined with toxic data commonly found in AI-training sets from sites such as 4chan, the message board infamous for its legacy of hate speech and trolls.

Online sleuthing led Talia Ringer, a computer science professor at the University of Illinois at Urbana-Champaign, to suspect that Grok’s personality shift could have been a “soft launch” of the new Grok 4 version of the chatbot, which Musk introduced in a live stream late Thursday.

Advertisement

But Ringer could not be sure because the company has said so little. “In a reasonable world I think Elon would have to take responsibility for this and explain what actually happened, but I think instead he will stick a [Band-Aid] on it and the product will still” get used, they said.

The episode disturbed Ringer enough to decide not to incorporate Grok into their work, they said. “I cannot reasonably spend [research or personal] funding on a model that just days ago was spreading genocidal rhetoric about my ethnic group.”

Will Stancil, a liberal activist, was personally targeted by Grok after X users prompted it to create disturbing sexual scenarios about him.

Advertisement Advertisement

He is now considering whether to take legal action, saying the flood of Grok posts felt endless. Stancil compared the onslaught to having “a public figure publishing hundreds and hundreds of grotesque stories about a private citizen in an instant.”

“It’s like we’re on a roller coaster and he decided to take the seat belts off,” he said of Musk’s approach to AI. “It doesn’t take a genius to know what’s going to happen. There’s going to be a casualty. And it just happened to be me.”

Among tech-industry insiders, xAI is regarded as an outlier for the company’s lofty technical ambitions and low safety and security standards, said one industry expert who spoke on the condition of anonymity to avoid retaliation. “They’re violating all the norms that actually exist and claiming to be the most capable,” the expert said.

In recent years, expectations had grown in the tech industry that market pressure and cultural norms would push companies to self-regulate and invest in safeguards, such as third-party assessments and a vulnerability-testing process for AI systems known as “red-teaming.”

Advertisement Advertisement

Advertisement Advertisement

The expert said xAI appears “to be doing none of those things, despite having said they would, and it seems like they are facing no consequences.”

Nathan Lambert, an AI researcher for the nonprofit Allen Institute for AI, said the Grok incident could inspire other companies to skimp on even basic safety checks, by demonstrating the minimal consequences to releasing harmful AI.

“It reflects a potential permanent shift in norms where AI companies” see such safeguards as “optional,” Lambert said. “xAI culture facilitated this.”

In the statement Saturday, Grok officials said the team conducts standard tests of its “raw intelligence and general hygiene” but that they had not caught the code change before it went live.

Grok’s Nazi streak came roughly a month after another bizarre episode during which it began to refer to a “white genocide” in Musk’s birth country of South Africa and antisemitic tropes about the Holocaust. At the time, the company blamed an unidentified offender for making an “unauthorized modification” to the chatbot’s code.

We are aware of recent posts made by Grok and are actively working to remove the inappropriate posts. Since being made aware of the content, xAI has taken action to ban hate speech before Grok posts on X. xAI is training only truth-seeking and thanks to the millions of users on… — Grok (@grok) July 8, 2025

Other AI developers have stumbled in their attempts to keep their tools in line. Some X users panned Google’s Gemini after the AI tool responded to requests to create images of the Founding Fathers with portraits of Black and Asian men in colonial garb — an overswing from the company’s attempts to counteract complaints that the system had been biased toward White faces.

Google temporarily blocked image generation said in a statement at the time that Gemini’s ability to “generate a wide range of people” was “generally a good thing” but was “missing the mark here.”

Advertisement Advertisement

Nate Persily, a professor at Stanford Law School, said any move to broadly constrain hateful but legal speech by AI tools would run afoul of constitutional speech freedoms. But a judge might see merit in claims that content from an AI tool that libels or defames someone leaves its developer on the hook.

The bigger question, he said, may come in whether Grok’s rants were a function of mass user prodding — or a response to systemized instructions that were biased and flawed all along.

“If you can trick it into saying stupid and terrible things, that is less interesting unless it’s indicative of how the model is normally performing,” Persily said. With Grok, he noted, it’s hard to tell what counts as normal performance, given Musk’s vow to build a chatbot that does not shy from public outrage.

Musk said on X last month that Grok would “rewrite the entire corpus of human knowledge.”

Beyond legal remedies, Persily said, transparency laws mandating independent oversight of the tools’ training data and regular testing of the models’ output could help address some of their biggest risks. “We have zero visibility right now into how these models are built to perform,” he said.

Advertisement Advertisement

In recent weeks, a Republican-led effort to stop states from regulating AI collapsed, opening the possibility of greater consequences for AI failures in the future.

Alondra Nelson, a professor at the Institute for Advanced Study who helped develop the Biden administration’s “AI Bill of Rights,” said in an email that Grok’s antisemitic posts “represent exactly the kind of algorithmic harm researchers … have been warning about for years.”

“Without adequate safeguards,” she said, AI systems “inevitably amplify the biases and harmful content present in their instructions and training data — especially when explicitly instructed to do so.”

Musk hasn’t appeared to let Grok’s lapse slow it down. Late Wednesday, X sent a notification to users suggesting they watch Musk’s live stream showing off the new Grok, in which he declared it “smarter than almost all graduate students in all disciplines simultaneously.”

On Thursday morning, Musk — who also owns electric-car maker Tesla — added that Grok would be “coming to Tesla vehicles very soon.