J. Robert Oppenheimer's regret came after Hiroshima and Nagasaki, once the weapon he built had already been used twice on cities full of people. He spent the rest of his life talking about what he'd helped unleash. That's the version of the story everyone knows: the brilliant physicist who saw what he'd made and couldn't unsee it, but only once it was too late to take back.
In February 2026, something that looks like an earlier version of that moment happened inside two of the world's most powerful AI labs, not after the fact, but in the middle of it. Two researchers didn't wait for a Hiroshima. They resigned instead, in public, warning about what they were watching happen from the inside.
Whether that counts as the same moment Oppenheimer had, or something meaningfully different from it, is the question worth sitting with.
Oppenheimer said the words out loud, after the fact. These two said them out loud, in the middle of it, and the work continued anyway.
Two resignations, two days apart, one shared warning
On February 9, 2026, Mrinank Sharma, who had led Anthropic's Safeguards Research Team since 2023, posted his resignation letter publicly. He wrote that "the world is in peril," not only from AI, but from a series of interconnected crises he saw unfolding, and that he had repeatedly struggled to let stated values actually govern decisions under commercial pressure. Two days later, Zoë Hitzig resigned from OpenAI in a New York Times guest essay, triggered by the company's decision to test advertising inside ChatGPT. She warned that monetizing a system people confide their medical fears and relationship problems to would repeat, in her words, the mistakes Facebook made, before anyone had built the tools to govern it safely.
Neither resignation was about a single reckless act. Both were about a pattern the researchers say they watched from the inside: institutional values that were real on paper losing out to competitive and commercial pressure in the moments that actually mattered.
Warnings from insiders carry a different weight than warnings from critics
Outside critics of AI have been predicting harm for years. Labs have largely been able to treat that criticism as the cost of doing business, expected, priced in, easy to wave off as people who don't understand the technology or the tradeoffs. Sharma and Hitzig are a different category of witness. They had access, credibility, and a stake in these companies succeeding. Their warnings didn't come from outside the room. They came from people who had been trusted to sit inside it.
That's what makes this closer to an Oppenheimer moment than ordinary industry criticism. Oppenheimer's authority came from having built the thing himself. Sharma's and Hitzig's authority comes from having watched, up close, how the decisions actually got made. Both kinds of warning carry weight precisely because the person delivering them had nothing to gain and a career to lose by saying it.
It's also not an isolated pattern. In 2024, OpenAI dissolved its Superalignment safety team after two of its leaders resigned, one of whom, Jan Leike, said he had been disagreeing with leadership about core priorities until reaching a breaking point. The February 2026 departures happened as OpenAI simultaneously dissolved a seven-person mission alignment team and, according to reporting at the time, removed a safety executive who had objected to a separate product rollout. Insiders leaving and naming the reason has become a recurring event, not a single anomaly.
The difference that actually matters isn't the warning. It's the timing
Oppenheimer's regret arrived after the irreversible act. Sharma and Hitzig delivered theirs before anything comparable to Hiroshima, while the systems they're worried about are still being built and shipped. That's either the more hopeful version of this story or the more damning one, depending on what happens next.
The hopeful reading: unlike 1945, there's still time to act on the warning. Nothing about this technology has reached a point of no return the way a detonated weapon has. The more damning reading: the warning already happened, publicly, from credible insiders, months ago, and the industry's response so far has mostly been to keep building at the same pace. If a warning this specific, from people this close to the work, doesn't change the trajectory, it's worth asking what kind of warning would.
There's a version of this pattern that predates AI entirely. Every consequential technology seems to produce its own internal dissenters eventually, the engineer who flags the flaw before the launch, the scientist who raises the ethical question before the deployment. What varies isn't whether the warning comes. It's whether the institution around it is built to actually hear it, or built to route around it and keep shipping. Sharma and Hitzig didn't lack an audience. They had one of the largest audiences a resignation letter has ever received. What they may have lacked is an institution structurally capable of changing course in response, regardless of how loud or credible the warning turned out to be.
So here's the question that doesn't resolve neatly: Oppenheimer is remembered as a man who saw the cost of what he'd built only once it was too late to matter. Sharma and Hitzig saw it, and said so, while it still might. History will eventually judge which kind of moment this actually was, the one where the warning came too late to change anything, or the one where it came in time and got ignored anyway. Right now, nobody outside those labs can tell which one we're living through.