Elon Musk wants AI bots to deliver the news to you. They're struggling with the task.

The limitations of the artificial intelligence model were exposed in the hours after the assassination attempt on Saturday, when the model displayed several incorrect headlines after reading X's content.

One headline erroneously reported that Vice President Kamala Harris had been shot, an error that appeared to stem from a sarcastic reference by some X users to an unrelated past incident in which President Biden had confused President Trump's name with Ms Harris'.

Another headline on Glock incorrectly named the alleged shooter, claiming the man was a member of Antifa, a loose network of far-left activists. Authorities have since given the suspect a different name, and a motive has yet to be determined.

Grok's articles are essentially jokes, rumors, and confusion elevated to breaking news on the platform, but beneath the summary is a disclaimer that reads, “This article is a summary of X's posts and may change over time. Grok may make mistakes. Please review the output.”

Musk has touted the benefits of his social media platform, X, using Grok's AI capabilities to automatically create headlines and news summaries based on hundreds of millions of users' posts. The billionaire, who has said traditional news media is slow and unreliable, has been encouraging X users for months to get into the habit of checking Grok for news updates.

“What we're doing with the X Platform is aggregation. We're using AI to bring together aggregated input from millions of users,” Musk said at an advertising industry gathering in June. “I think this is really going to be a new model for news.”

Company X did not respond to a request for comment on Grok's headline.

Grok is a product of Musk's artificial intelligence company xAI, which began development on it last year to provide chatbots and other features to X subscribers.

The technology could be promising, as Grok has accurately summarized some news, and Musk said an advantage is that his AI model has access to X's real-time posts. One of Grok's unique features is that it's what X calls an “AI search assistant with a humorous twist.”

These errors suggest that design can also be a weakness, highlighting the potential pitfalls of trying to make a humor-oriented computer model sort through a large number of posts from a variety of sources in real time.

“We still have a long way to go,” said Katie Harbath, a former public policy director at Facebook. “At the end of the day, when it comes to breaking news like a shooting, you always need human help to provide context when the facts aren't yet known.”

Journalists sometimes make mistakes after surprising and dramatic events like the attacks on Trump, but some of Grok's missteps went far beyond the confusion that can come with such a moment.

Another Grok headline read, “Home Alone 2 Actor Shot at Trump Rally?” Trump made a cameo appearance in the 1992 film Home Alone 2, a reference made by some X-users. Grok did not specify that the “actor” in question was Trump. The summary read, “An actor from the film Home Alone 2 was reportedly shot during a public event presumed to be a Trump rally.”

Grok isn't the first generative AI tool to have accuracy issues: In another example, Google said in May it was making fixes after an AI-powered feature produced strange results, such as recommending the use of glue to make cheese stick to pizza.

Some companies have taken a different approach from Musk and haven't positioned their products as geared toward real-time news. Asked about the shooting on Monday, OpenAI's ChatGPT provided a summary of the incident and linked to fact-checking site PolitiFact, but also offered a disclaimer that it is “not a real-time news product.” News Corp, owner of The Wall Street Journal, has a content licensing partnership with OpenAI.

Meta Platforms executives have said they have no plans to promote political content on its Thread platform, a microblogging site that competes with X. Meta executive Adam Mosseri said last year that the platform wants to avoid the negativity and other downsides that often accompany politics and hard news.

Some users on the thread complained over the weekend that the platform was slower to release information about the shooting than X, suggesting that Musk's platform remains a resource for news consumers.

X Chief Executive Officer Linda Yaccarino said in an email to employees on Sunday that X was treating the assassination attempt as a “serious incident” and was prioritizing resources to “take swift action” against posts that violate the platform's rules, such as those glorifying the attack, according to a copy of the email obtained by The Wall Street Journal.

This isn't the first time Grok has struggled to summarize a news event. After the June presidential debate, Grok appeared to stumble when an X user mentioned California Governor Gavin Newsom, creating the headline “Newsom wins recent debate.” Newsom did not participate in the debate but did attend the event. Some political commentators have pegged Newsom as a possible successor to Biden, though Newsom has publicly rejected the idea.

Twitter previously had a team that manually wrote summaries of trending topics. Musk disbanded that team after the acquisition as part of larger staff cuts, said Evan Hansen, a former Twitter executive who worked on those efforts until the cuts. Musk renamed the platform “X” after buying Twitter in 2022.

Hansen said Twitter had experimented with using AI to write summaries before Musk bought it, but still needed humans to review them. “AI can work, but it has to be very clever,” he said.

Write to Alexa Corse at alexa.corse@wsj.com.