Elon Musk’s Grok chatbot has started talking “gibberish” to users.

“App scale designs with that and to your horse,” the system told one user, according to reports on Reddit. “For consider best many times data, it the to to red issues of plastic slow due,” it said to another.

Those are just two of a run of reports from users that the system had apparently started talking nonsense to users who had seemingly asked normal questions.

Using Grok’s official X account, developers xAI said that the strange results were an accident.

“That pure word salad is a rare temporary generation glitch,” it said. “Start a fresh chat or regenerate—it usually clears right away.

“Sorry about the gibberish.”

Grok is built by xAI, Elon Musk’s artificial intelligence company that was founded in 2023. It is built into the social network X, where many users interact with it by posting to it, but it also has a devoted website of its own.

Since it launch three years ago as an explicitly less controlled competitor to other chatbots such as OpenAI’s ChatGPT and Anthropic’s Claude, Grok has run into a series of controversies. It has been found to praise Adolf Hitler and allow users to create sexualised images of children.

xAI was founded as a separate artificial intelligence company as part of X, the company that runs the social network previously known as Twitter. Earlier this year, it was bought by SpaceX – ahead of an IPO of the combined company this summer, which made Elon Musk the world’s first trillionaire.The Independent is the world’s most free-thinking news brand, providing global news, commentary and analysis for the independently-minded. We have grown a huge, global readership of independently minded individuals, who value our trusted voice and commitment to positive change. Our mission, making change happen, has never been as important as it is today.