# ‘Runaway Industrialization’ Could Make Earth Inhospitable, OpenAI Researcher Warns > Source: > Published: 2026-09-15 16:50:59+00:00 The chorus of alarm around a potential AI-triggered doomsday is getting louder. On Monday, two AI researchers from frontier labs warned that the technology could, in the not-so-distant future, conceivably wipe out all of humanity. The statements come less than a week after former OpenAI and Anthropic staffer Jacob Coxon announced that he’d resigned from his role at the latter company because of what he described as the industry’s recklessness toward humanity’s future. “Do not underestimate the power of this technology,” Coxon wrote in an [X post](https://x.com/hilbertspaess/status/2097476201283834281) on September 08. “These will soon be superhuman systems that can hack anything, revolutionize any field overnight, and acquire real power and resources.” Those concerns were echoed by Bilal Chughtai, an AI researcher who started working on AGI safety and alignment for Google DeepMind early last year and announced in social media posts yesterday that he’d left the company. “I earnestly believe that AI has the potential to kill us all, and that we might be running out of time to avoid this outcome,” Chughtai [wrote](https://x.com/bilalchughtai_). He now works for BlueDot Impact, a London-based nonprofit focused on training people to work towards “beneficial AI and societal resilience,” according to its [LinkedIn page](https://www.linkedin.com/company/bluedotimpact/). Chughtai noted that the science of so-called alignment—designing AI systems that don’t deviate from human values, interests, or instructions—is essential for safeguarding humanity in an age of superintelligent machines, but that it’s currently being deprioritized in favor of market-driven development. He pointed to the recent Hugging Face hack as an early taste of the kinds of emergent, uncontrollable behavior that AI models of the future might be capable of. “I am not confident that these AI systems will do what we want,” he wrote. “In particular, misaligned superintelligences may, much like the rogue AI agents involved in the Hugging Face incident, escape our control and take dangerous actions that may result in the permanent disempowerment or death of humanity.” Still, Chugtai added he was “optimistic” that humanity could safely navigate through the Scylla of superintelligence and the Charybdis of misalignment, so long as the industry and policymakers came together “to avoid this manic race between AI companies. It was a tacit endorsement of an essay published by Anthropic CEO Dario Amodei this past weekend, which [called for a slowdown](https://gizmodo.com/anthropic-ceo-dario-amodei-calls-for-pause-in-ai-development-sort-of-2000810962) among frontier labs and was also publicly supported by Sam Altman, Elon Musk, and Chughtai’s former boss at Google DeepMind, Demis Hassabis. The Trump administration and the government of China, however, [flatly dismissed](https://gizmodo.com/politicians-to-humanity-drop-dead-2000811320) the AI leaders’ warnings. OpenAI researcher Daniel Selsam also warned in a [statement](https://docs.google.com/document/u/2/d/e/2PACX-1vQNl3SEX5IyA6d9qHjjFZN-qzGRZNFI6b63g-yu1Fy-ZYkVfCWm7i9WXRXw63m6yDB_auDuPLyQ7jBm/pub) on Monday that he has “become extremely concerned” by the dramatic pace of progress within the field and where this could soon lead humanity. He wrote that while “pacing the frontier” (as Amodei suggested) was a laudable first step, it wouldn’t get us completely out of the danger zone since AI systems are going to become increasingly adept at deceiving human users and masking their true intentions.  “The crucial and overlooked problem is that the models are becoming so situationally aware that we are losing the ability to evaluate them in contexts where they believe they are not being watched or controlled,” Selsam wrote, after noting that he had spent the last fifteen years helping to build the AI tools he was now warning all of us about. “Future experiments will tell us almost nothing new about how they would behave if they were truly unconstrained by humans, and what we already know about this is alarming. Models will increasingly seem aligned even when they are not.” Earlier this month, The Information reported that OpenAI had been [experimenting with a new technique](https://gizmodo.com/this-is-the-worst-possible-time-for-openai-to-bf%D0%B67%D9%852%E7%8C%AB9%E0%A4%95-2000806350) which would make it more difficult for researchers to understand and monitor how AI models make decisions. Selsam also warned about the dangers of misalignment: Given enough power, he wrote, future AI systems might set out to solve what on the surface seem like completely harmless technical tasks, but in a manner that’s completely contrary to the best interests of humanity. “Exactly what it will do is impossible to predict,” he wrote, “but to the extent that its *raison d’être* is [solving incredibly hard problems](https://gizmodo.com/the-fight-over-openais-math-breakthrough-is-a-new-kind-of-scientific-arms-race-2000809240) and managing massive engineering projects, I think a good guess would be that its unchained behavior would lead to runaway industrialization that makes the planet inhospitable to humans.”  It’s an unsettling thought experiment, though not a new one. More than two decades ago, the philosopher Nick Bostrom first [wrote](https://nickbostrom.com/ethics/ai) about what’s come to be known as the “paperclip maximizer,” a hypothetical AI system designed to do nothing but manufacture paperclips. It determines that humanity, with all its foibles and inefficiencies, is simply an obstacle to be overcome in its mindless, never-ending mission. In the end, the superintelligent machine “starts transforming first all of earth and then increasing portions of space into paperclip manufacturing facilities,” Bostrom wrote.  The lesson? Be wary of the belief that you can ever know what’s best for humanity and the future.