More than two years since Generative AI swept the world after its public offering on Chatgpt, trust has been a permanent issue.

Hallucinations, bad mathematical and cultural biases plague the results, reminding users that at least for the time being there is a limit to the extent we can rely on AI.

Elon Musk's Grok chatbot, created by his startup XAI, shows this week that there is a deeper reason to focus: humans are easily manipulated by humans.

Grok began responding to user inquiries on Wednesday to false claims of “white genocide” in South Africa. By the end of the day, the screenshot posted a similar answer on X even though the question had nothing to do with the topic.

After 24 hours of silence on the matter late Thursday, Groke's strange behavior was caused by the so-called "unauthorized modification" of the chat app's so-called system prompts, which helps inform them of their behavior and how they interact with users. In other words, humans are deciding how AI responds.

In this case, the nature of the reaction is directly linked to Musk who was born and raised in South Africa. Musk, in addition to serving as CEO Tesla SpaceX has been pushing for false claims that violence against some South African farmers constitutes "white genocide", which is also the point President Donald Trump has expressed.

"It's very important because of the content and the content that leads the company, and the way it suggests or articulates the power of the tools that must shape people's thinking and understanding of the world," said Deirdre Mulligan, a professor at the University of California, Berkeley.

Mulligan describes Grok Miscue as a "algorithm decomposition" that "tears at seams" the neutral nature of the large language model. She said there is no reason to consider Groke's failure as an "exception."

AI-powered chatbots are Yuan,,,,, Google Instead of "packaging" information in a neutral way, Openai passes data through a set of built-in filters and values, Mulligan said. Grok's breakdown provides a window to easily change these systems to meet the agenda of an individual or group.

Representatives from XAI, Google and Openai did not respond to requests for comment. Mehta declined to comment.

Xai said in its statement that Groke's unauthorized changes violated "internal policies and core values." The company said it will take steps to prevent similar disasters and will release system prompts for the app to "strengthen your trust in Grok as a truth-seeking AI."

This is not the first AI error for online viruses. Ten years ago, Google's photo app mistakens African Americans for gorillas. Last year, Google temporarily suspended its Gemini AI image generation feature after acknowledging it provided "inaccuracy" in historical images. Some users accused OpenAI's DALL-E image generator of showing signs of bias in 2022, which led the company to announce that it is implementing a new technology, so images "accurately reflect the diversity of the world's population."

Forrester found that 58% of AI decision makers in the UK and US companies expressed concern about the hallucination risks of generative AI deployment in 2023. The survey in September that year included 258 respondents.

Experts told CNBC that the Grok incident is reminiscent of China's DeepSeek, which became a sensation in the U.S. overnight due to the quality of its new model, reportedly at a small part of the cost of its U.S. competitors.

Critics say DeepSeek examiners are sensitive to topics in China's government. They say that like China has a deep wet China, Musk seems to influence the results based on his political views.

When Xai made its debut in November 2023, Musk said it would have been to be "a bit of a witty" and "a rebellious winning streak" and answered "spicy questions" that rivals might evade. In February, Xai accused an engineer of a change that curbed answers to users’ questions about misinformation, making Musk and Trump’s names unreply.

But Groke's recent obsession with "white genocide" in South Africa is even more extreme.

Petar Tsankov, CEO of AI model audit firm Latticeflow AI, said the Grok explosion was more surprising than what we saw with DeepSeek because people “want to do some sort of manipulation from China.”

Tsankov, whose company is based in Switzerland, said the industry needs to be more transparent so that users can better understand how companies build and train models and how they affect behavior. He noted that the EU's efforts to require more technology companies to provide transparency as part of the region's broader regulations.

Tankov said without strong public outcry, “we will never be able to deploy a safer model, and it will be that “people who will pay” put their trust in the companies that develop these companies.

Forrester analyst Mike Gualtieri said the Grok crash is unlikely to slow down the growth of chatbots’ users or reduce the investment in companies’ influx of the technology. He said that users have a certain degree of acceptance of these events.

"Whether it's Grok, Chatgpt, or Gemini - now everyone expects it," Gualtieri said. "They were told how the model was hallucinating. Expect this to happen."

AI ethicist and author of AI’s book, Olivia Gambelin, published last year, said that while Grok’s activity may not be surprising, it highlights fundamental flaws in the AI ​​model.

"It is at least possible to use the Grok model, and you can adjust these general base models at will," Gambelin said.

- CNBC's Lora Kolodny and Salvador Rodriguez contribute to this report

watch: Elon Musk's Xai Chatbot Grok put forward the "white genocide" claim in South Africa.