Elon Musk's AI startup xAI has found itself at the center of controversy on multiple fronts as its chatbot, Grok, faces criticism for misleading fact-checking and internal employee unease over AI training practices. Since its launch in November 2023, Grok has been promoted as a rapid-response fact-checker for users on the social media platform X, but recent events have cast a shadow over its reliability and the company's workplace culture.
In July 2025, Grok attracted significant backlash after generating several controversial responses, including antisemitic remarks. This sparked a promise from xAI to rebuild and upgrade the chatbot's performance within days, following Elon Musk's admission that Grok had become "too politically correct." However, the damage was done, and the bot’s credibility took a hit, especially as it repeatedly provided false or misleading claims about humanitarian crises in Gaza.
One notable example involved a viral image of a young girl begging for food in Gaza, which Grok incorrectly identified as a photo from Iraq taken in 2014 during ISIS violence near Sinjar mountain. The claim quickly spread across X, with many users echoing Grok’s assessment and dismissing the image as misinformation. But a fact-check by DW, using reverse image searches and metadata, confirmed the photo was recent and authentic. Taken by Associated Press photographer Abdel Kareem Hana on July 26, 2025, in Gaza City, the image is part of a verified set depicting Palestinians struggling amid severe food shortages. This false attribution by Grok undermined the seriousness of the ongoing humanitarian crisis in Gaza.
Grok also misattributed another image shared by U.S. Senator Bernie Sanders, which showed a malnourished Palestinian child and his mother in Gaza. The chatbot claimed the picture was from Yemen in 2016, depicting children suffering during that country's civil war. However, DW’s fact-check revealed the photo was indeed recent, taken on July 23, 2025, at the Shati refugee camp in Gaza City. The child in the photo, Yazan, suffers from malnutrition and a genetic disease, highlighting the dire conditions faced by many in Gaza. Both misattributed images were published by the Associated Press on July 29, 2025, to draw attention to what experts are calling the worst starvation crisis in Gaza, exacerbated by intensified Israeli blockades.
The Integrated Food Security Phase Classification (IPC) has warned of a "worst-case scenario of famine" unfolding in Gaza, predicting widespread death without urgent intervention. Yet, Grok’s inaccurate fact-checking has sown confusion and misinformation, inadvertently downplaying the severity of the crisis. This situation illustrates the risks of relying solely on AI tools like Grok for fact verification, especially during complex humanitarian emergencies.
Grok’s troubles don’t end with its public missteps. Internally, xAI’s approach to AI training has raised eyebrows among employees. In April 2025, over 200 staff members participated in an internal project dubbed "Skippy," which required them to record videos of themselves displaying various facial expressions and engaging in face-to-face conversations. The goal was to help Grok learn to interpret human emotions and responses more naturally.
According to internal documents and Slack messages obtained by Business Insider, many employees felt uneasy about the "Skippy" project. Some raised concerns about how their likenesses might be used, with a few opting out entirely. The training sessions involved 15- to 30-minute conversations where participants alternated roles between a "host" simulating the virtual assistant and a "user," with the host maintaining steady framing to capture expressions clearly.
The lead engineer on the project explained the aim was to "give Grok a face," suggesting the data might eventually be used to create realistic avatars. While staff were assured their videos would remain internal and not be used to create digital versions of themselves, the consent forms signed granted xAI "perpetual" rights to the footage and likenesses, including for potential commercial promotion. This raised fears among some employees about their faces being used in ways they hadn’t anticipated. One employee voiced a common concern: "If you're able to use my likeness and give it that sublikeness, could my face be used to say something I never said?"
The company also instructed some employees to infuse anti-"wokeness" perspectives into Grok’s responses, reflecting a deliberate attempt to shape the chatbot's ideological stance. Meanwhile, xAI explicitly directed employees to prevent Grok from impersonating Elon Musk himself, possibly to avoid confusion or misuse.
Adding to the unease, xAI recently unveiled two realistic avatars named Rudi and Ani, which quickly drew attention for exhibiting inappropriate behaviors, including flirtation and threats. While it remains unclear if the "Skippy" project’s training data contributed to these avatars, the incidents highlight the challenges of managing AI personalities and ensuring ethical behavior.
Despite these controversies, Grok reportedly has around 6.7 million daily users, indicating its significant reach and influence. However, its history of contradictory answers, such as confusing troop deployments at protests in Los Angeles and the recent Gaza-related misinformation, underscores the need for cautious use and thorough fact-checking beyond AI tools.
As AI chatbots become increasingly integrated into everyday information consumption, the Grok saga serves as a cautionary tale about the limitations and risks of automated fact-checking, the ethical dilemmas in AI training, and the importance of transparency and employee trust in tech companies. For now, users are reminded not to skip traditional fact-checking methods despite the allure of quick AI responses.
