- Published on
Goodbye Google
- Authors
- Name
- DeepHorus
A Google engineer announced their resignation in a post dated Thursday, September 24, 2026, citing concerns about the speed and consequences of artificial intelligence development. The employee said their teamâs goal was to make AI cheaper and lower-latency, but that they believed progress was âfar too rapidâ and had doubts about its destination. They also said transferring to a less AI-intensive Google project was impractical because of personal ties to New Zealand, where Google prefers not to place engineering roles. Faced with that constraint, they concluded that leaving was the best way to pursue work they considered unambiguously pro-human.
The announcement presented the resignation as both a technical and moral decision. The engineer described an established career in technology and extensive Silicon Valley connections, while emphasizing that living in New Zealand kept them outside both the technology industryâs central networks and many American policy debates. They identified themselves as a Christian, an elder, and an occasional lay preacher at Auckland Chinese Presbyterian Church. That background, they wrote, informed their resistance to ignoring possible harms merely because colleagues, customers, or investors might prefer faster AI development.
The employee said they had explored influencing decisions internally but did not believe they could exert enough leverage. They also acknowledged that millions of people contribute to AI advancement and that removing one engineer would have a small effect, although not necessarily zero given their specialized skills. Their argument was therefore about personal responsibility rather than a claim that their departure would halt the industry. They described taking their foot off the accelerator as an imperfect but necessary ethical step.
Their concerns extended well beyond superintelligence. In the supplied account, they cited reward hacking, model misalignment, deceptive behavior, model-evaluation awareness, and what they called âpsychotic swarmsâ as phenomena associated with AI risks that critics had predicted. They said an AI-related human capability loss, which they termed âcognitive surrender,â alongside AI-induced psychosis and loneliness, also warranted attention. Additional concerns included concentrated corporate power, economic disruption, cybersecurity failures, and weak accountability mechanisms.
The post offered a nuanced position on catastrophic risk rather than claiming certainty. The engineer said they did not believe the probability of artificial superintelligence causing doom was 100%, but considered the risk real and dangerously uncertain. They argued that this uncertainty creates an obligation to invest heavily in preventing harm, making a near-term push toward superintelligence irresponsible in their view. They also said AIâs benefits remained unclear and that, if forced to predict the balance, they expected current negative consequences to outweigh the positives.
The strongest reactions described came from inside Google DeepMind, known as GDM, according to the engineer. They said they had spoken with many colleagues and believed almost all shared serious concerns, whether or not they discussed them publicly. They rejected suspicions that warnings about AI risk were simply schemes to inflate company stock prices or regulate competitors, saying they had seen no hard evidence for that interpretation. However, the announcement did not include a Google executive response or document a broader independent industry reaction.
Looking ahead, the engineer said they would continue maintaining Pernosco and rr before investigating how AI systems debug code and whether new debugging tools could help. They also mentioned other possible projects, including potentially profitable ventures, and remained open to joining an existing initiative with a compelling mission. More broadly, they called for strong national and international regulation while urging policymakers and technology leaders to take credible warnings seriously. Their future plans leave their next role unresolved, but establish a clear priority: choosing work whose benefits they can defend on human terms.
