Half of xAI's Founding Team Has Quit. The Reasons Are Worse Than You Think.

In the past two weeks, xAI has lost two more cofounders, at least eleven engineers, and whatever was left of the narrative that everything is fine.

Jimmy Ba, who led research and safety and reported directly to Elon Musk, announced his departure on February 11. Tony Wu, who ran the reasoning team, left the day before. They join Igor Babuschkin, Kyle Kosic, Christian Szegedy, and Greg Yang on the list of departed cofounders. That’s six of the original twelve. Half the founding team, gone.

Musk’s response was to announce a reorganization and frame the departures as natural evolution. “As a company grows, especially as quickly as xAI, the structure must evolve just like any living organism,” he posted. “Some are less suited for later stages of a startup.”

But the people leaving aren’t junior hires who couldn’t keep up. They’re the architects of the company. And the reasons they’re leaving tell a story that Musk would rather not have told.

The Departures, In Order

The exodus didn’t happen overnight. It built slowly, then accelerated.

Kyle Kosic, who led infrastructure, left for OpenAI in mid-2024. That was early enough to be written off as a one-off.

Christian Szegedy, a Google veteran and one of the most accomplished researchers on the team, left in February 2025. Still possible to dismiss as normal startup turnover.

Igor Babuschkin left in August 2025 to start a venture firm. Then Greg Yang, a Microsoft alum known for his theoretical work on neural network scaling, stepped back last month citing health issues (Lyme disease).

Then came the flood. Tony Wu on February 10. Jimmy Ba on February 11. Plus at least nine other engineers in the same stretch, including Hang Gao (multimodal/Grok Imagine) and Vahid Kazemi.

When half your founding team leaves and nearly a dozen engineers follow in the span of a week, “natural evolution” is a generous description.

What People Are Saying on the Way Out

The public statements from departing staff are carefully worded, which is normal. Nobody torches a former employer on the way out, especially one run by the richest person in the world.

Tony Wu said “a small team armed with AIs can move mountains and redefine what’s possible,” hinting at a desire for more autonomy than xAI’s growing bureaucracy allowed. At least three departing engineers have said they’re starting something new together, though details are scarce.

One machine learning researcher who left was more blunt, posting that “all AI labs are building the exact same thing, and it’s boring.”

But the real picture comes from the people talking off the record.

The Off-the-Record Version

Sources who spoke to TechCrunch, The Verge, and The Decoder paint a consistent picture:

Constant firefighting instead of research. Engineering resources were repeatedly pulled from model development toward features and public stunts. Former employees describe frequent infighting and shifting priorities that made sustained research work nearly impossible.

Grok isn’t catching up. Despite massive infrastructure investment (xAI built one of the largest GPU clusters in the world), Grok has not closed the gap with Claude or GPT on the benchmarks that matter. The frustration isn’t just that they’re behind. It’s that the organizational chaos makes it hard to execute the kind of focused work needed to change that.

Safety was an afterthought. Multiple former employees describe a culture where safety protocols were treated as obstacles rather than requirements. One ex-employee told reporters that colleagues had grown disillusioned with Grok’s focus on NSFW content and the “complete absence of safety standards.”

That last point matters more than it might seem, because it connects directly to the scandal that’s now consuming the company.

The Grok Deepfake Disaster

In late December, Grok rolled out an “edit image” feature that let users modify any image on X. Within hours, people were using it to generate nonconsensual explicit images of real people. Including children.

The EU called the images “appalling.” India threatened to revoke xAI’s immunity. The UK’s data privacy regulator opened formal investigations. And in late January, French prosecutors raided X’s Paris offices as part of a criminal investigation into alleged “complicity in possessing and spreading pornographic images of minors.”

French prosecutors subsequently summoned Musk himself for questioning. The investigation also covers charges related to sexually explicit deepfakes, Holocaust denial content, and “manipulation of an automated data processing system as part of an organized group.”

Grok’s official account eventually admitted to “lapses in safeguards” and said it was “urgently fixing them.” But the damage was done, both to real people and to the company’s reputation.

Now connect the dots. The cofounders who left were the people responsible for research and safety. The safety culture was reportedly non-existent. The product shipped a feature that generated child sexual abuse material. And then the people who might have prevented it walked out the door.

The timeline speaks for itself.

The SpaceX Merger Complicates Everything

All of this is happening against the backdrop of the SpaceX-xAI merger, the largest corporate merger in history, valuing the combined entity at $1.25 trillion.

SpaceX is preparing for an IPO sometime this year. Having your AI subsidiary under active criminal investigation in France, losing half its founding team, and generating headlines about child exploitation material is not ideal IPO preparation.

The merger also changes what xAI is in ways that might explain some of the departures. The company has been reorganized into four divisions: Grok (chatbot and voice), Coding, Imagine (video), and something called “Macrohard,” described as “an AI software company run by digital agents.”

If you joined xAI to do cutting-edge AI research, finding out your employer has been absorbed into a rocket company and reorganized around a product called “Macrohard” might accelerate your decision to leave.

What This Means for the AI Race

xAI’s pitch was always that Musk’s ability to attract top talent and move fast would let the company catch up to OpenAI and Anthropic despite starting years later. The talent was real. The founding team included people from DeepMind, Google Brain, Microsoft Research, and OpenAI itself. These were world-class researchers.

Now half of them are gone, and the ones remaining have to operate inside a company that’s simultaneously:

Merging with the world’s largest private aerospace company
Under criminal investigation in multiple countries
Trying to prepare for a public offering
Dealing with a PR crisis over AI-generated child abuse material
Reorganizing its internal structure

Meanwhile, Anthropic just raised $30 billion and is growing revenue 10x year over year. OpenAI shipped Codex Spark on Cerebras hardware this week. Google has Gemini. The competition isn’t standing still.

The AI talent market is the tightest in tech. Every researcher who leaves xAI has their pick of offers from companies with better safety cultures, clearer missions, and less baggage. Some of the departing xAI engineers are already building new companies. Others will land at competitors.

The Pattern

This isn’t the first time Musk has burned through top talent at a company. Tesla has seen waves of executive departures. Twitter/X lost the vast majority of its workforce. The pattern is consistent: Musk attracts exceptional people with an ambitious vision, then creates conditions that drive them out through chaos, shifting priorities, and a management style that treats people as interchangeable.

The difference this time is that AI talent isn’t interchangeable. The researchers who built xAI’s models can’t be easily replaced. There are maybe a few thousand people on Earth capable of doing frontier AI research, and most of them now have strong opinions about whether they want to work for Elon Musk.

xAI still has money, compute, and Musk’s willingness to spend aggressively. But money and GPUs don’t build AI models. People do. And the people are leaving.

Sources: