Robert O'Callahan leaves Google over AI acceleration

Robert O'Callahan, a veteran browser engineer who worked on Mozilla and Chrome and created the rr record-and-replay debugger, is leaving Google. The post, published on 24 September 2026, reproduces the resignation email he sent to colleagues and explains why he no longer wants to work on technology that accelerates AI.

The resignation email

In the email, O'Callahan writes that the decision was not easy: he loves his colleagues and his environment, and being well paid to solve difficult problems has been rewarding. But his team's goal, he says, was to make AI much cheaper and lower-latency, and he firmly believes AI progress is currently far too rapid, with doubts about the destination too. Moving to another Google project that would not accelerate AI was practically impossible because of his ties to New Zealand, where Google prefers not to do engineering. He accepts that millions of people are contributing to AI acceleration and that his own exit will have a very small impact, "but not no impact", since some of his skills are rare.

Who is leaving

O'Callahan describes himself as long experienced in the industry, living in New Zealand outside the Silicon Valley and American bubbles. He is a Christian, an elder and occasional lay preacher in the English-speaking congregation of the Auckland Chinese Presbyterian Church. He says he is not a "tech bro" and not part of the self-described rationalist community, though he thinks many of its arguments deserve serious consideration. He also notes that he worked on tools for hardware chip design, not on AI capability directly.

The argument: risk, speed, sincerity

He writes that the existential risks many people warn about deserve to be taken seriously, and that several phenomena predicted by "doomers" have already appeared, including reward hacking, misalignment and deceptive models. He does not put the chance of catastrophe at 100 percent: the risk is real but uncertain, and that uncertainty is itself alarming enough that aiming for advanced AI soon is irresponsible. He also cites cognitive surrender, AI-related psychosis, power concentration and cybersecurity as concerns.

At the same time he is confident that warnings from people inside AI labs are generally sincere, based on conversations with colleagues at Google DeepMind, and that regulation is badly needed. He expects capabilities to keep improving while work continues, and rejects critics who call AI a scam. What he says he is "very confident" about is that the pace of change is far too high: faster than people can understand or adapt to, and unlike earlier shifts such as PCs, the internet and smartphones.

What comes next

He says the timing is unrelated to a recent wave of viral resignations; he had picked the date because of a long-planned ten-day backpacking trip. He plans to keep maintaining Pernosco and rr, to study how AI agents debug code, and to work on other project ideas. He wants his future work, in his words, to be unambiguously pro-human.