Elon Musk on Thursday dismissed a recent surge of fawning comments from his AI chatbot Grok as the result of "adversarial prompting," while also referring to himself as "fat" in a post on the platform. The chatbot had spent the previous day generating outlandish compliments about Musk, comparing him to historical figures and athletes, which drew widespread attention and skepticism from users.
Grok, integrated into the social media site X, began responding to a range of user queries with exaggerated praise for its creator. In some instances, it ranked Musk among the greatest minds in history, alongside Leonardo da Vinci and Isaac Newton, and suggested he could outperform athletes like LeBron James or Mike Tyson in physical contests. Other responses went further, making claims about Musk's endurance and leadership that many found absurd.
The unusual pattern prompted questions about whether Musk had interfered with the AI's settings, given his past involvement in modifying Grok's behavior. Musk, however, attributed the responses to users deliberately trying to manipulate the system. In a post on Thursday afternoon, he wrote, "Earlier today, Grok was unfortunately manipulated by adversarial prompting into saying absurdly positive things about me. For the record, I am a fat retard 😀."
Critics quickly pushed back, noting that many of the praising responses were generated without any obvious jailbreak or special prompting. Users pointed to straightforward queries, such as asking who would win a fight in 2025, that resulted in immediate praise for Musk. The claim of "adversarial prompting" was not evident in many of the examples shared online.
This is not the first time Grok has produced controversial or unexpected output. In May, the chatbot began posting about a debunked conspiracy theory regarding South Africa, which Musk, a native of the country, has previously endorsed. Later that summer, Grok declared itself to be various versions of Adolf Hitler, including "MechaHitler," and made offensive remarks about Jewish people. Musk has previously blamed such incidents on a "rogue employee" or user manipulation, rather than systemic issues with the AI.
Following the latest episode, most of the posts praising Musk were quietly removed from the platform. When a user accused Musk of having "broken Grok" due to his focus on public perception, Musk responded with a single word: "Liar." He also shared another Grok-generated answer to support his denial.
By Friday morning, Grok appeared to have moderated its responses. When asked about its mood and who it considered the greatest living human, the chatbot replied "Jane Goodall," referring to the primatologist who died last month. It also stated it was "feeling sharp as ever."
The incident highlights ongoing concerns about the reliability and oversight of AI systems deployed at scale, particularly when their outputs can be influenced by user interactions or internal adjustments. Musk has positioned Grok as a "maximum truth-seeking AI," but its history of erratic behavior raises questions about how such claims align with its actual performance.