Meta CEO Mark Zuckerberg at a sporting event on March 8, 2025 in Las Vegas, Nevada. (Photo: Cooper Neill/Zuffa LLC/Getty Images)Cooper Neill/Zuffa LLC/Getty Images

Meta has reportedly delayed the rollout of its flagship AI model after engineers struggled to substantially improve its capabilities.

Dubbed “Behemoth,” the large language model was beaten to market in April by lesser models in Meta’s Llama AI family. 

The flagship model was initially delayed to June; now, Behemoth is expected in the “fall or later,” according to the Wall Street Journal. 

Meta has never publicly committed to a date.

The slipping timeline is enough to raise “internal concerns about the direction of its multibillion-dollar AI investments,” according to the Journal. (Meta didn’t comment.) 

The company plans to spend some $72 billion in capital expenditures this year, most of it committed to AI projects.

Meta may very well release Behemoth earlier than that by limiting its capabilities. But the bigger issue is why the slide happened in the first place. 

“Senior executives at the company are frustrated at the performance of the team that built the Llama 4 models and blame them for the failure to make progress on Behemoth,” notes the Journal. (What happens next? To paraphrase Karen O: Heads could roll.)

Still, Meta’s hardly alone in its challenges. Rivals like OpenAI have also run into capability limitations for their flagship AI models as engineers reach the upper limit of what’s possible using the current approach. —AN

If you were using the social media service X on Wednesday, you may have noticed something…strange going on.

The Grok chatbot embedded in the Elon Musk-owned service, which is powered by Musk’s xAI, kept steering every conversation to the topic of “white genocide” and South Africa.

The context wasn’t lost on those who read the news. Days prior, the Trump administration welcomed white South Africans to the U.S. as refugees. Musk, of course, has led White House efforts to reduce the size and cost of the federal government. And Musk was born in Pretoria.

Still, the question remained: How could this happen? 

xAI offered answers the following day.

“An unauthorized modification was made to the Grok response bot's prompt on X,” the company wrote in a social media post. “This change, which directed Grok to provide a specific response on a political topic, violated xAI's internal policies and core values.”

Moving forward, xAI said it would share its Grok system prompts on GitHub for public review and add “additional checks and measures” to require review of prompt modifications.

Not that any of this stopped OpenAI CEO Sam Altman, who has famously fallen out with Musk and whose company competes with xAI, from taking a dig.

“There are many ways this could have happened. I’m sure xAI will provide a full and transparent explanation soon,” he wrote. “But this can only be properly understood in the context of white genocide in South Africa …” —AN

Coinbase said in a regulatory filing Thursday that it will incur expenses of $180 million to $400 million in the wake of a cyberattack that breached the account data of a "small subset" of its customers.

An unknown entity sent the cryptocurrency exchange an email on May 11 claiming to have customer information and internal documents.

Coinbase corroborated the claim—no login credentials or passwords were obtained, but names, addresses, and emails were—but refused to pay a $20 million ransom. 

Working with law enforcement, the company determined that the hackers had paid multiple Coinbase workers, staffers and contractors alike, to collect information. It quickly dismissed them. 

In a twist, Coinbase has created a $20 million reward for information on the perpetrators.

Disclosure of the hack—plus an SEC probe into the accuracy of the company’s user figures—comes shortly before Coinbase joins the benchmark S&P 500 index, a milestone for the crypto category. —AN