There’s something darkly poetic about the head of AI safety at one of the world’s most influential AI companies quitting his job to pursue… actual poetry.
On February 9th, Mrinank Sharma posted his resignation letter on X. Sharma led Anthropic’s Safeguards Research Team — the people responsible for making sure Claude doesn’t help you build a bioweapon or manipulate you into buying crypto. His letter reads like a tech worker having an existential crisis at a silent retreat, complete with Zen quotes and William Stafford verses.
“The world is in peril,” he wrote. “And not just from AI, or bioweapons, but from a whole series of interconnected crises unfolding in this very moment.”
Then he announced he’s off to get a poetry degree.
The Context That Makes This Spicy
Let’s be clear about timing here. Sharma’s resignation comes days after Anthropic’s Claude Cowork launch sent shockwaves through Wall Street. The new AI model, with its enhanced plugins and agent capabilities, spooked investors so badly that tech stocks took a nosedive. The fear? That Claude is now capable enough to automate meaningful chunks of white-collar work — particularly in legal and administrative roles.
Even more telling: internal surveys revealed Anthropic employees themselves are worried. “It kind of feels like I’m coming to work every day to put myself out of a job,” one staffer confided. “In the long term, I think AI will end up doing everything and make me and many others irrelevant,” said another.
So the guy in charge of making sure AI doesn’t harm humanity decided to bounce right when his own colleagues are terrified that their creation might hollow out the job market. Interesting timing, no?
Reading Between the Vague Lines
Sharma’s letter is frustratingly non-specific, which is either professional courtesy or cowardice depending on your read. But there are breadcrumbs.
“Throughout my time here, I’ve repeatedly seen how hard it is to truly let our values govern our actions,” he wrote. “I’ve seen this within myself, within the organization, where we constantly face pressures to set aside what matters most.”
Translation: Anthropic talks a big game about safety, but when push comes to shove — when the competitive pressure heats up, when investors want results, when OpenAI launches something new — those principles get squishy.
This is the same company that positions itself as the “responsible” AI lab. The one founded by ex-OpenAI researchers who worried their former employer wasn’t taking safety seriously enough. The company that published lengthy AI safety cases and blogs about constitutional AI. If even they are facing pressures to “set aside what matters most,” what hope does the rest of the industry have?
The Departure Pattern
Sharma isn’t leaving in isolation. Last week, at least two other notable Anthropic employees — researcher Harsh Mehta and AI scientist Behnam Neyshabur — announced departures to “start something new.” Meanwhile, Dylan Scandinaro, another former Anthropic AI safety researcher, just jumped ship to become… head of preparedness at OpenAI.
Yes, you read that correctly. Anthropic’s safety person went to OpenAI. The company that Anthropic’s founders literally left because they thought it wasn’t safe enough.
Either the AI safety field is one big game of musical chairs where principles are flexible, or something is shifting at these labs that’s making the safety-conscious folks uncomfortable.
The Poetry Escape
Here’s what gets me about Sharma’s exit: he’s choosing to become “invisible.” His words. He wants to study poetry, practice “courageous speech,” and explore “poetic truth alongside scientific truth.”
There’s something both admirable and damning in that choice. Admirable because he’s clearly done some genuine soul-searching about what matters. Damning because it suggests he sees no path forward within the system. The guy who was literally paid to make AI safe has concluded that the most authentic thing he can do is… leave.
It’s worth noting that Sharma cited a book advocating for something called “CosmoErotic Humanism.” One of its authors is Marc Gafni, a controversial figure with documented accusations of manipulating and exploiting followers. That’s a strange intellectual rabbit hole for an AI safety researcher to tumble down, and it raises questions about how much of this resignation is genuine concern versus personal crisis dressed up in philosophical language.
What This Actually Means
Let me give you the pessimistic reading: AI safety as practiced at major labs is fundamentally compromised. The field attracts earnest researchers who genuinely want to prevent catastrophic outcomes. But those researchers exist within corporate structures that have raised billions of dollars on the promise of building powerful AI systems. The incentives are misaligned. You can talk about safety all you want, but when you’re in an arms race with OpenAI and Google, when your next funding round depends on shipping impressive capabilities, when your employees are privately worrying they’re building the tools that will replace them — safety becomes the thing you compromise on.
The optimistic reading: Sharma is one person having a personal crisis. Anthropic is still shipping safety research. Claude still refuses to help with dangerous tasks. The system, imperfect as it is, is working well enough.
I land somewhere in the middle. I think Anthropic genuinely cares more about safety than most of its competitors. I also think that caring more than the competition is an embarrassingly low bar. And I think when your safety lead writes a resignation letter that sounds like a cry for help wrapped in poetry quotes, maybe it’s time to ask harder questions about what’s happening behind the scenes.
The Uncomfortable Truth
The AI industry has a safety theater problem. Labs publish papers about alignment and red-teaming and constitutional AI. They hire smart people to work on these problems. They position themselves as the responsible ones. And then they ship models that make their own employees fear for their jobs.
Mrinank Sharma looked at this situation — the competitive pressures, the race dynamics, the gap between stated values and actual behavior — and decided the most authentic response was to leave and write poetry.
That’s not reassuring. That’s alarming.
The question isn’t whether AI will change everything. It clearly will. The question is whether anyone with the power to steer these systems is actually steering, or whether they’ve all quietly resigned themselves to being passengers.
“You don’t ever let go of the thread,” ends the William Stafford poem that Sharma included in his letter.
I wonder: at Anthropic, at OpenAI, at all these labs racing toward superintelligence — who’s still holding the thread? And is it enough?
DevDigest Now is your daily dose of tech analysis. We cut through the PR speak to tell you what actually matters. No sponsors, no fluff, just straight talk about the industry that’s reshaping everything.