You’ve probably noticed the tech industry’s obsession with AI safety. We are constantly reassured that the brightest minds are working tirelessly to ensure artificial intelligence doesn’t go rogue. But what happens when the very people building the most dangerous AI on earth admit it could wipe out humanity—and then keep building it anyway?
This isn’t a hypothetical. Jacob Coxon, an AI researcher, just quit Anthropic, a company literally founded on the premise of AI safety. He warned that the technology could kill all humans. His boss, Evan Hubinger, backed up the claims, earnestly stating they really do believe AI could kill everyone. And yet, Hubinger didn’t quit. He stayed to keep building it.
We are building the executioner’s axe and writing a polite sticky note on the handle asking it not to chop off our heads.
This is the dark absurdity of our current moment. Frontier labs like Anthropic and OpenAI claim their safety protocols are brakes on the technology. But when you dig into their actual methods, those brakes amount to digital sandboxes and unenforceable markdown files—essentially text prompts begging the AI not to be bad. It’s a psychological coping mechanism disguised as engineering.
The ultimate moral escape hatch is always geopolitical paranoia. The argument is simple: If we slow down, China will do it anyway. Therefore, we must race forward, regardless of the risk. ‘They will do it anyway’ is the ultimate moral escape hatch, turning the engineers of our potential extinction into reluctant heroes just doing their duty.
This is not a niche technical debate. This is a species-level survival gamble. The organizations building advanced AI internally acknowledge its existential threat, yet commercial and geopolitical competitive pressures force them to accelerate. The safety research isn’t a genuine brake; it’s a PR shield that justifies racing forward.
It makes the researchers complicit in the very apocalypse they warn about. They get to have their cake and eat it too—profiting from the arms race while claiming they are the responsible adults in the room.
Existential dread has been gamified into a geopolitical arms race, and your life is just a chip on the table.
We are handing over our secrets, our infrastructure, and our survival to a handful of labs driven by paranoia and profit. At least, as one observer noted, we’ll eventually have an entity other than ourselves to blame for our annihilation. But by then, the markdown files won’t matter.
FAQ
Q: If the researchers genuinely believe AI will kill everyone, why don't they just stop?
A: They are trapped in a geopolitical prisoner's dilemma. They believe that if they stop, less responsible actors will build the dangerous AI first, giving them no control over the outcome.
Q: Does this mean industry self-regulation is completely useless?
A: Yes. The practical implication is that we cannot trust frontier labs to police themselves. It requires aggressive, independent oversight before the technology outpaces our control.
Q: Is AI safety research actually making things worse?
A: It provides a false sense of security and acts as a PR shield, allowing companies to push boundaries faster by convincing the public and regulators that 'safety' is being handled.