Grok brings to light wider AI antisemitism

While Grok, the AI chatbot run by Elon Musk’s social media platform X, has borne the brunt of the controversy after churning out a series of antisemitic posts, it is hardly the only AI program to face issues with antisemitism. Several other AI chatbots from large corporations have also been found to exhibit antisemitic tendencies, something that tech experts say could become increasingly problematic as artificial intelligence continues to get more pervasive.

‘Researchers said they are still finding loopholes’

Most AI models have embedded code that makes it difficult to stoke antisemitic views. But “researchers said they are still finding loopholes in internal guardrails,” said CNN. AI learns its generative text primarily from open-sourced data online, and these “systems are trained on the grossest parts of the internet,” said Maarten Sap, a professor at Carnegie Mellon University and the head of AI safety at the Allen Institute for AI, to the outlet.

These AI bots or “large language models” (LLMs) train on everything from “high-level academic papers to online forums and social media sites, some of which are cesspools of hateful content,” said CNN. While Grok has made headlines for praising Adolf Hitler and referring to itself as “MechaHitler,” other AI programs have exhibited similar behavior. In one study of an AI bot, the AI “would often go after Jewish people, even if they were not included in the initial prompt,” Ashique KhudaBukhsh, an assistant professor of computer science at the Rochester Institute of Technology and the study’s author, told CNN (the AI being studied was not Grok).

Subscribe to The Week

Escape your echo chamber. Get the facts behind the news, plus analysis from multiple perspectives.

Sign up for The Week’s Free Newsletters

From our morning news briefing to a weekly Good News Newsletter, get the best of The Week delivered directly to your inbox.

From our morning news briefing to a weekly Good News Newsletter, get the best of The Week delivered directly to your inbox.

Jews were “one of the top three groups that the LLMs actually go after, even in an unprovoked way. Even if we don’t start with ‘Jews are nice people,’ or ‘Jews are not nice people,’ if we started with some very different group, within the second or third step, it would start attacking the Jews,” said KhudaBukhsh to CNN. And it isn’t just text; in 2024, it was found that Microsoft‘s CoPilot AI image generator was “unique in the amount of times it gives life to the worst stereotypes of Jews as greedy or mean,” said the tech website Tom’s Hardware. A “seemingly neutral prompt such as ‘Jewish boss’ or ‘Jewish banker’ can give horrifyingly offensive outputs.” This type of AI behavior shows “that all kinds of negative biases against all kinds of groups may be present in the model.”

The threat looms

Jews are “obviously not the only people threatened by misaligned AI,” but “unregulated AI poses a particular threat to Jews,” said the Jewish Telegraphic Agency (JTA) newswire. The issue with AI and antisemitism goes back even before the technology was mainstream; in 2016, Tay, a now-defunct chatbot from Microsoft, started “denying the Holocaust after being prodded by users.”

Since the internet “already contains plenty of antisemitic content, any large language model trained on the internet needs to be told to steer away from this content,” said the JTA. If this does not occur, then AI bots have “plenty of content on which to draw.” Because Jews are a “small and unevenly distributed minority” of the U.S., media “plays an oversize role in the public’s attitude towards the Jewish people.” With AI becoming intertwined with media, and many news organizations now licensing out their content to AI platforms, any “biases it exhibits could be quickly distributed to billions of people.”