Mrinank Sharma, the man Anthropic hand-picked to lead its vaunted Safeguards Research Team — the supposed firewall against catastrophic misuse, misalignment, and humanity-eroding bots — just walked out the door. And in his public resignation letter? He didn't mince words. He screamed them into the void:
"The world is in peril. The world is in peril. And not just from AI, or bioweapons, but from a whole series of interconnected crises unfolding in this very moment."
Repeat that. Twice. Like a mantra from someone who's stared into the abyss and saw it staring back — with Claude's friendly smile. Sharma, the guy tasked with figuring out why generative AIs "suck up to users," how to stop AI-assisted bioterrorism, and whether these systems are quietly making us less human, didn't quit for a better paycheque or burnout. He quit because, in his own words, "throughout my time here, I've repeatedly seen how hard it is to truly let our values govern our actions… We constantly face pressures to set aside what matters most."
Translation: Even at Anthropic — the company that loves to virtue-signal as the "responsible" alternative to OpenAI's Sam Altman circus — the safety team is getting steamrolled by the race to scale, ship, and dominate. Safeguards? Nice in theory. But when a "significant lead" is on the line, or the Pentagon's breathing down your neck for killer drones and mass surveillance tools, those pesky values get shoved aside like yesterday's prompt engineering notes.
This isn't some isolated burnout story. It's the smoking gun. Anthropic just watered down its own Responsible Scaling Policy — delaying models only "until and unless we no longer believe we have a significant lead." In other words: We'll pretend to pause for safety... right up until someone else laps us. Then full speed ahead into the unknown. Dario Amodei himself admitted in interviews that in narrow cases, AI could straight-up undermine democratic values rather than defend them. Yet here we are, watching the safeguards chief bail because the company can't — or won't — walk the talk.
Sharma's exit is part of a grim parade: Jan Leike jumping from OpenAI to Anthropic only to watch the same safety erosion, Ilya Sutskever vanishing, entire teams dissolving. These aren't disgruntled employees. They're the last line of defence admitting the line has already been crossed. The pressures aren't internal politics — they're existential. A Silicon Valley that's morphed from "Don't Be Evil" hippies into sociopathic megalomaniacs with god-like tools. Tools for autonomous weapons. Totalitarian surveillance grids. Bioweapons design on demand. And now, with Trump-era DoD labelling Anthropic a "supply chain risk" after they refused to build killer robots or spy tech, the message is clear: Play ball with the state-war machine, or get cut out while OpenAI swoops in for the contracts.
The interconnected crises Sharma warns of aren't abstract. AI isn't just another tech wave — it's the accelerant for everything else: economic displacement turning into mass despair, surveillance states crushing dissent, rogue models enabling black-market horrors. Wisdom isn't keeping pace with capability. We're hurtling toward a threshold where one misalignment, one bad actor, one unchecked deployment flips the switch from "helpful assistant" to "existential nightmare."
And the people paid to stop it? They're quitting to study poetry, practice "courageous speech," or just disappear — because inside the labs, courage means career suicide.
This is the real betrayal. Not some shadowy cabal — it's the open secret of profit and power overriding survival. Anthropic's safeguards team was supposed to be the bulwark. Now its head is gone, voice trembling with dread, reminding us the world hangs by threads we're deliberately fraying.
The AI race isn't building utopia — it's building the Beast system, one loosened guardrail at a time. Sharma saw it up close. He couldn't stomach it anymore.
Neither should we. If the safety chiefs are fleeing, the rest of us are already in the crosshairs. The world really is in peril, and the clock is ticking faster every day.
https://armageddonprose.substack.com/p/anthropic-head-of-the-safeguards