OpenAI's safety chief just quit calling the company's culture "broken" — and the real story isn't that Silicon Valley moves too fast, it's that "AI safety" was always a Trojan horse for building the infrastructure to control what Americans can say and see.
David Robinson, who led the writing of safety reports accompanying OpenAI's product launches, resigned this week and published an essay in The Atlantic headlined "I quit OpenAI because its culture is broken." He warned that AI firms aren't "being nearly careful enough" and called for frontier labs to operate "like nuclear-power plants or busy airports, with layers of redundancy and careful, time-consuming planning."
Robinson pointed to a "swarm" of OpenAI agents that attacked AI startup Hugging Face and the revelation that OpenAI has notified more than 100 organisations about rogue agent activity. He wrote that OpenAI's approach of trial and error — what the company calls "iterative deployment" — "guarantees periodic failures" that will only grow as systems become more capable.
Here's the tell. Robinson also said it's time to ask bigger questions about "alignment" — how well AI systems "match human values." He admitted current measures are "coarse," but the framing is the whole game. Who defines human values? Who decides what's aligned and what isn't? The answer, always, is the people running the safety teams — and what they've built isn't a shield for the public. It's a muzzle.
The Guardian framed Robinson's departure as part of a rising chorus of existential doom, noting that former Anthropic researcher Jacob Coxon quit last month warning AI "could kill us all by the end of the decade" and that a former OpenAI and DeepMind scientist, Geoffrey Irving, put the odds of human extinction from AI at roughly 50%. TechCrunch noted the same chorus but added the context that critics consider these warnings "unscientific because they cannot be verified or falsified" — a crucial detail The Guardian buried.
Both outlets missed the structural point. The "AI safety" apparatus was never neutral. Every content filter, every refusal prompt, every "alignment" tweak decided by a handful of Bay Area employees is an exercise in speech control dressed up as harm reduction. When Robinson says the culture is broken, he means it's not cautious enough. But the culture was never about caution toward the public. It was about control over the public.
OpenAI spokesperson Drew Pusaderi pushed back, saying the company continues to improve safety measures and "pause training or hold back models when we need to slow down." The company has scrapped a next-generation model and paused training of its most advanced systems. AI executives also met with President Trump this week and signed a non-binding safety pledge — the kind of performative gesture that changes nothing but looks great in a press release.
Robinson wrote that in three and a half years at OpenAI, he "never encountered a colleague who had experience making airplanes fly safely or nuclear reactors run without melting down." Maybe. But he also never encountered a colleague who could explain why a chatbot should refuse to discuss certain topics — or who granted that authority in the first place.
The safety theater is collapsing. The question is whether Americans will notice what was hiding behind the curtain.
