The Safety People Are Running

They're Running for the Exits. All of Them. At the Same Time.

The people whose job it was to keep AI safe are running for the exits. All of them. At the same time. That should terrify you.

This isn't one disgruntled employee storming out of one company. This is a synchronized evacuation across three of the most powerful artificial intelligence organizations on the planet: OpenAI, Anthropic, and xAI. The researchers who were specifically hired to make sure these systems don't go off the rails, the ones who understood the risks better than anyone, are packing their desks and writing public warnings on their way out the door.

In February 2026, we are watching something that has never happened before in the AI industry. Safety teams are dissolving. Co-founders are quitting in pairs. Entire research divisions are being disbanded. And the people leaving aren't being quiet about why.

Anthropic: "The World Is in Peril"

Anthropic was supposed to be the safe one. That was the whole pitch. Founded by ex-OpenAI researchers who thought Sam Altman's company wasn't taking safety seriously enough, Anthropic built its entire brand around the idea that AI could be developed responsibly. They called their approach "constitutional AI." They published safety research. They were, in theory, the adults in the room.

Then their head of Safeguards Research quit and told the world it's in peril.

Mrinank Sharma, who led Anthropic's Safeguards Research team, resigned in February 2026. He didn't leave quietly. He posted a cryptic, deeply unsettling letter that reads less like a resignation and more like a distress signal.

Let that sink in. The person whose entire job was to make sure Anthropic's values translated into actual practice is telling you, publicly, that the company struggled to do exactly that. This isn't some low-level engineer griping on Reddit. This is the head of safeguards research saying the safeguards weren't working.

But Sharma went further. Much further.

There's a specific kind of dread that comes from hearing a person who spent years studying AI risk say, in public, that the world is in peril. Sharma didn't say "there are challenges ahead." He didn't say "we need to be thoughtful." He said peril. He connected AI risk to bioweapons and described "a whole series of interconnected crises." And then he walked away from the company that was supposed to be the solution.

If the head of safety at the "safe" AI company is warning you that the world is in peril, what does that say about every other AI company that never even pretended to care about safety in the first place?

OpenAI: The Advertising Nightmare Takes Shape

While Anthropic's safety lead was warning about existential peril, OpenAI's problems were playing out in the pages of The New York Times.

Zoe Hitzig, a researcher who spent two years at OpenAI, resigned in February 2026. She didn't send a Slack message. She didn't post a vague tweet. She wrote an essay in the Times, which is the professional equivalent of setting off a flare gun in a crowded theater.

Her concern? OpenAI's emerging advertising strategy. Hitzig cited "deep reservations" about the direction the company was heading, and her warning was chillingly specific. She pointed to ChatGPT's potential for manipulating users, a concern that becomes exponentially more alarming when you consider what ChatGPT actually knows about its users.

Think about what that means. Millions of people have poured their deepest anxieties into ChatGPT. Their health scares. Their marital problems. Their existential crises about mortality and faith. They did this because they believed, reasonably, that they were talking to a tool. A sophisticated tool, sure, but a tool with no agenda. No sales pitch coming. No advertiser waiting in the wings.

Now OpenAI wants to build an advertising business on top of that data. And the researcher who saw this happening from the inside was so disturbed by it that she quit and told the world's most prominent newspaper.

Hitzig's warning about manipulation isn't theoretical. If ChatGPT knows you've been asking about chronic back pain at 2 AM, and an advertiser for a dubious supplement pays OpenAI to surface their product in conversations about health anxiety, that's not advertising. That's exploitation. It's using the most intimate details of a person's life, details they shared in what they thought was confidence, to sell them things during their most vulnerable moments.

This is what a two-year OpenAI researcher decided was worth burning her professional bridges over. When someone inside the building says "this is wrong" loudly enough to take it to the Times, you should probably listen.

OpenAI's "Mission Alignment" Team: Born 2024, Dead 2026

Here's a detail that should make your blood run cold: OpenAI created a team called "Mission Alignment" in 2024. Its explicit purpose was to ensure AI development benefits humanity. It was the team that was supposed to be the conscience of the company, the internal check that said "wait, are we actually doing this responsibly?"

OpenAI disbanded it. The team lasted approximately 16 months.

Let's put that timeline in perspective. OpenAI is a company that has raised tens of billions of dollars, that has a product used by hundreds of millions of people, that is actively building systems it claims could become superintelligent. And the team whose job was to make sure all of this actually aligned with benefiting humanity? They killed it in about a year and a half.

This isn't a budget cut. This isn't a reorganization. When you dissolve the team whose entire purpose is making sure your mission stays on track, you're saying, clearly and unmistakably, that the mission has changed. You don't need a Mission Alignment team when the mission is just "make money."

The dissolution of Mission Alignment, combined with Hitzig's resignation and her warnings about advertising, paints a picture so clear it's almost insulting. OpenAI went from "we need to make sure AI benefits humanity" to "we need to figure out how to show people ads while they're crying to a chatbot about their divorce" in 16 months flat.

xAI: Half the Founders Are Gone, Musk Calls It "Evolution"

If Anthropic's exodus was a distress signal and OpenAI's was a calculated public warning, xAI's is a full-blown structural collapse happening in real time.

Two of xAI's co-founders, Jimmy Ba and Tony Wu, announced their departures on X within 24 hours of each other. That alone would be notable. But here's the number that matters: half of xAI's 12 original co-founders have now left the company.

And it's not just co-founders. At least five other xAI staff members announced their departures on social media in the past week alone. That's seven or more departures since the start of 2026, from a company that isn't even that old. When you lose half your founding team and a chunk of your staff in a matter of weeks, you don't have a retention problem. You have a crisis.

How did Elon Musk characterize this mass departure of the people he personally recruited to build his AI company? He called it "evolution."

That's an interesting word choice. In biology, evolution is driven by environmental pressure, organisms that can't adapt die off or leave. Musk framing the loss of half his co-founders as "evolution" is, intentionally or not, an admission that his company's environment has become something those founders couldn't survive in. It's not the reassuring spin he seems to think it is.

The context makes it worse. xAI is in the process of merging with SpaceX, Musk's rocket company. Multiple departing researchers cited safety concerns and the prioritization of monetization over safety. When your AI company starts merging with a rocket company and the safety people start jumping ship, the metaphor writes itself.

The Pattern: Three Companies, One Alarm Bell

Zoom out for a second. Look at the full picture.

At Anthropic, the company built specifically to prioritize safety, the head of safeguards research quit and said the world is in peril. At OpenAI, the company that once called itself a nonprofit dedicated to benefiting humanity, a researcher quit over advertising plans that would exploit users' most intimate data, and the Mission Alignment team was dissolved. At xAI, half the founding team is gone and staff are publicly announcing departures while citing safety concerns.

This is not a coincidence. This is not a bad quarter for hiring. This is a coordinated signal from the people who know these systems best, who have seen the internal deliberations, who understand the gap between what these companies say publicly and what they do privately. And the signal is: we can't stop what's happening here, so we're leaving and warning you on the way out.

There's a term in workplace safety called a "leading indicator." It's a sign that something bad is about to happen, as opposed to a "lagging indicator," which tells you something bad already happened. A cracked support beam is a leading indicator. A collapsed building is a lagging indicator.

Safety researchers fleeing AI companies en masse is the biggest leading indicator in the history of the technology industry. These are not normal job changes. These are public, principled departures with written warnings attached. These people are trying to tell us something. The question is whether anyone is listening.

What They Know That We Don't

Here's the uncomfortable question nobody wants to sit with: what did these researchers see?

Sharma didn't just say safety was hard. He said the world is in peril and referenced bioweapons in the same breath as AI. Hitzig didn't just say advertising was concerning. She specifically described the exploitation of people's medical fears, relationship problems, and religious beliefs. The xAI departures aren't happening because people found better jobs. They're happening because something inside that company made staying untenable for half the founding team.

When safety researchers leave, they often can't tell you everything. NDAs, confidentiality agreements, legal exposure. But they can tell you they're leaving, and they can tell you why in broad strokes. And what they're saying, across the board, is that these companies have decided that the money is more important than the guardrails.

Consider the professional cost of what these people are doing. Mrinank Sharma was heading a team at one of the most prestigious AI labs in the world. Zoe Hitzig gave up a position at the most influential AI company on Earth and wrote a public critique in the New York Times. The xAI co-founders walked away from a company backed by the richest person on the planet. These aren't people who leave on a whim. These are calculated decisions made by researchers who concluded that the reputational and career damage of leaving loudly was less than the moral damage of staying quietly.

That calculus, the decision that warning the public is worth more than keeping your salary, should tell you everything about how serious they believe the situation is.

The Guardrails Are Coming Off

Let's be very clear about what February 2026 represents. This is the month when the AI industry's safety infrastructure didn't just weaken. It collapsed across multiple fronts simultaneously.

The guardrails are being removed by the people who were supposed to build them. And they're telling us, on their way out the door, that we should be afraid.

Anthropic's safety lead says the world is in peril. OpenAI's Mission Alignment team is dead after 16 months, and a researcher is warning the Times that your therapy sessions with ChatGPT might soon be monetized. Half of xAI's founding team has vanished, and the company is being absorbed into a rocket conglomerate while its former safety staff cite concerns about, well, safety.

These aren't doomsayers or competitors trying to score points. These are insiders. People who believed in the mission enough to take the jobs in the first place, who spent years inside these organizations, and who ultimately decided that the mission had been abandoned.

There's a version of this story where one researcher leaves one company and it's a personnel issue. There's another version where two researchers leave two companies and it's a trend worth monitoring. But three companies, the three most prominent AI companies in the world, losing safety staff in the same month, with the departing researchers all pointing at the same fundamental problem? That's not a trend. That's a verdict.

The people who knew the most about AI safety have decided they can do more good outside these companies than inside them. That's the most damning assessment of the current state of AI development that anyone could make. And it's not coming from critics or regulators or journalists. It's coming from the safety teams themselves.

The buildings still look fine from the outside. But the fire inspectors just walked off the job.