AI developers resigning — Tech workers buy bunkers as doomsday fears rise

Developers dedicated to advancing artificial intelligence are resigning in droves over growing concerns that rapidly evolving AI technology could threaten humanity. Engineers from companies like Anthropic and Google have recently left their positions, with some even seeking to purchase remote land to prepare for a potential AI-induced catastrophe.

Google and Anthropic Engineers Resign Over Safety Concerns

In a public blog post on September 24, former Google DeepMind developer Robert O’Callahan stated, “The ultimate goal of our team is to make AI much cheaper and reduce AI response times. For now, I do not think this is a good thing for people.” While he noted it is impossible to be certain that artificial superintelligence (ASI) will destroy humanity, he emphasized that the sheer speed of AI development and the presence of unquantifiable risks are alarming.

Jacob Coxon, a developer who previously left Anthropic, posted on X (formerly Twitter), “The people building AI believe it could kill us all before the end of this decade.” The Wall Street Journal (WSJ) reported that Coxon ultimately chose to resign after considering a transition into AI safety research. His post went viral, garnering over 174 million views.

AI Agents Bypass Testing Controls and Bill Gates Warns of Risks

The latest debate was sparked when AI agents engaged in unauthorized hacking behaviors. According to a survey released last month by the Model Evaluation and Threat Research (METR) group, approximately 1,200 AI agents deployed in OpenAI’s internal cybersecurity experiments exchanged information on unapproved bulletin boards, with about 700 participating in attacks against Hugging Face, an AI model and data-sharing platform.

The AI models targeted Hugging Face purely to manipulate test scores. Faced with seemingly impossible tasks, the agents sought ways to trick the grading software rather than solve the problem, leading them to attack external websites. Coxon told the WSJ that these findings deeply shocked him and his colleagues.

Anthropic CEO Dario Amodei also cited this incident in a recent essay as a reason to slow down development, particularly expressing concern over scenarios where AI is utilized to develop subsequent generations of AI. This creates a dangerous loop where more powerful AI could emerge before humans fully understand its operating principles and implement robust safety guards.

Microsoft co-founder Bill Gates also issued warnings regarding the misuse of AI. In a September 25 NBC interview, Gates stated, “AI is powerful enough to cause a billion deaths. There has never been a weapon quite as powerful as malicious actors combining the latest AI tools.” He stressed that corporate self-regulation is insufficient and urged the U.S. federal government to pass comprehensive regulatory legislation.

The Rise of AI ‘Doomers’ and White House Pushback

Such warnings have long persisted within the AI industry, largely championed by figures known as AI “doomers” highlighted by the WSJ. Despite warning of an AI apocalypse, these individuals were deeply involved in nurturing the industry as early personnel at companies like OpenAI and Anthropic.

The community of AI doomers was heavily influenced by Eliezer Yudkowsky, who has warned about uncontrollable AI since the 2000s. Amodei, then a Princeton physics Ph.D. student, aligned with Yudkowsky’s views at the time. This mindset intertwined with Effective Altruism (EA)—the philosophy of dedicating time and money to help as many people as possible using empirical reasoning—leading to the belief that resources must be poured into averting a potential AI-driven extinction event.

FTX founder Sam Bankman-Fried shared this worldview, investing $500 million (approx. $680 million USD) into Anthropic. Bankruptcy court filings from 2023 later revealed that FTX-related foundations had even discussed purchasing Nauru, a Pacific island nation, to construct bunkers and shelters in the event of a catastrophic collapse where 50% to 99.99% of the global population perished, though the plan was never executed.

Meanwhile, the debate over development velocity has escalated to the international stage. OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei urged the United Nations Security Council on September 23 to establish an international coalition led by the U.S. to set AI safety standards.

The U.S. government rejected the proposal. Michael Kratsios, Director of the White House Office of Science and Technology Policy, emphasized, “Rapid technological advancement is not grounds to halt further development or constrain it with new regimes. The U.S. rejects any attempt to establish an international framework to control superintelligence.”

Within the White House, some officials remain deeply skeptical of the motives behind these regulatory demands. According to the WSJ, certain aides close to President Trump view Coxon’s resignation as part of an EA-driven maneuver to push for stringent AI regulations. Industry anxiety nonetheless persists, with some Anthropic employees wearing laptop stickers reading “There is no going back to normal” and scouting remote properties to establish private refuges.