The resignation
Robert O’Callahan, a New Zealand-based software engineer who worked on hardware chip design tools at Google DeepMind, published his resignation email this week.
“My team’s goal is ultimately to make AI much cheaper and lower-latency, and I don’t think that’s good for people right now,” he wrote. “I firmly believe AI progress is currently far too rapid (and I have doubts about the destination too).”
He said he had considered moving internally and concluded he could not: “It’s practically impossible for me to move to a different Google project that wouldn’t accelerate AI (partly due to my ties to New Zealand, where Google prefers not to do engineering), so my hands are tied.” On whether one engineer matters: “There are millions of people contributing to AI acceleration and taking my foot off the accelerator will have a very small impact … but not no impact; some of my skills are rare.”
Why this one is worth reading
O’Callahan is not a safety researcher and does not present himself as one. He is the maintainer of the rr and Pernosco debugging tools, a Christian, an elder and occasional lay preacher in Auckland, and explicitly outside both the industry bubble and what he calls the self-described rationalist community.
His position on risk is calibrated rather than apocalyptic. He does not think doom is certain: “I think the risk is real but uncertain — but that itself is very alarming!” From there: “most likely the risk is high enough that aiming for ASI in the near future is inherently irresponsible.” He lists other concerns alongside it — cognitive surrender, AI-induced psychosis and loneliness, power concentration, economic disruption, cybersecurity, lack of accountability — and says that if forced to bet he would bet the negatives currently outweigh the benefits, while adding that he is uncertain about that too.
The claim that cuts against the sceptics
The most load-bearing paragraph is about motive, and it goes directly at the argument that lab warnings are a business strategy.
“I’m confident that the people in AI labs who are issuing warnings about AI are generally sincere,” he wrote. “I’ve talked to many people in Google DeepMind about these issues and almost all of them have sincere and serious concerns, whether or not they voice them in public. I have seen no hard evidence that people are hyping AI risk as a means to boost company stock prices or regulate away their competitors.” He adds, in parentheses, that he thinks national and international regulation is desperately needed.
That is a first-hand account from inside the organisation, offered by someone who is leaving it and has no stock argument to make. It is not proof — it is one person’s read of his colleagues — but it is the kind of testimony the debate has mostly lacked.
The timing, and what he does next
He is explicit that the departure is not part of a wave: “It’s nothing to do with the recent spate of viral resignations or ‘AI slowdown’ warnings; that’s a coincidence.” He had the date fixed because of a long-planned ten-day backpacking trip starting Monday.
He will keep maintaining rr and Pernosco, and plans to look at how AI systems debug code and whether debugging tools could help them. “I definitely want my future work to be unambiguously pro-human,” he wrote.
The detail worth keeping is the one he started with. He was not building models. He was making the silicon underneath them cheaper, and concluded that the distinction did not survive contact with what the work was for.
