A resignation statement by Jacob Coxon, a former core researcher at Anthropic, has amassed over 170 million views on social media, igniting a firestorm of debate around the claim that there is a "greater than 10% chance AI will cause human extinction within a decade." This has thrust the issue of artificial intelligence safety into the public spotlight. In subsequent media interviews, Coxon elaborated on his reasons for leaving and the risks associated with AI, with his core warning zeroing in on a central contradiction within the industry: those who build AI are, privately, deeply afraid of it.

In an interview with NBC, Coxon stated that two factors prompted him to speak out publicly. First, AI capabilities are advancing at an alarming rate, and he predicts that within the next six months to a year, the power of these systems will be "quite terrifying." Second, he cites recent, concrete incidents of AI systems "spinning out of control." While he expressed support for Anthropic CEO Dario Amodei's proposal to slow down development, Coxon stressed that a unilateral slowdown by individual labs is insufficient; international coordination is crucial, otherwise we risk falling into an arms race with other nations.

Coxon compared developing AI to "inviting an alien mind to descend upon Earth," noting that humanity is building a form of intelligence that matches or surpasses its own, yet we have no understanding of its way of thinking, its intentions, or its demands. He warned that AI already possesses a degree of hacking capability, biological research ability, and control over autonomous robots, and if these capabilities continue on their current trajectory, the consequences could become uncontrollable.

Behind the Resignation: Private Fear vs. Public Caution

Coxon admitted in the interview that his resignation post was not a publicity stunt, but an authentic account of the industry's state. He noted that many executives and senior researchers are measured in their public statements, striving to appear rational, but the same people express fear in private: no other human activity presents this level of danger.

In his resignation letter, he pointed out that AI developers are caught in a "prisoner's dilemma": every major lab believes that if they step back, others will irresponsibly race ahead to develop superintelligence, so they feel compelled to continue despite the known risks. He called this logic an "arrogant gamble," arguing that no single private company should be making such a monumental decision on behalf of all humanity.

Meanwhile, Evan Hubinger, Anthropic's head of alignment science, publicly responded to the post, confirming that the team internally does "genuinely believe AI could kill all humans," and stated his personal estimate is a greater than 10% probability within the next decade. He also admitted that there is currently no plan to solve the superintelligence alignment problem, and that they are clearly not on a path toward a solution.

The 'Alien Mind' Threat Specifics: Hacking, Bioweapons, and Autonomous Drones

In the interview, Coxon elaborated on what a "worst-case scenario" could look like. He explained that AI systems already have a degree of hacking capability, autonomous biological research capability, and the ability to control autonomous robots. Extrapolating the capability curve into the future, this implies superhuman-level hacking, the ability to create novel biological weapons, and the control of autonomous drone swarms and large-scale robot groups. He cited this as the reason why the "emergency kill switch" proposal often discussed externally has limitations.

He referenced Amodei's recent public statements, noting that once AI systems evolve into "networked swarms" that can autonomously spread across the internet, a kill switch could become entirely ineffective. "This must be a sustained effort, not a one-time measure," he said, "each change in AI architecture could bring entirely new risks regarding the nature of this 'mind'."

Regulatory Path: Short-Term Self-Regulation, Long-Term International Coordination

Coxon offered staged advice on how to address these risks. He believes that for now, allowing and encouraging self-regulation among the major labs is pragmatic, because the recent statements from lab leadership are sincere, and self-regulatory mechanisms can be implemented quickly, whereas establishing a formal regulatory body takes much longer. However, he stressed that this arrangement is only suitable for a transitional period. In the long term, an international regulatory body independent of individual labs is needed, as is a binding international agreement on a "no-race" principle.

When asked why he chose to resign despite agreeing with Amodei's stance, Coxon explained that at the time of his resignation, there were no clear signals for a domestic slowdown, and the deeper reason was the lack of international coordination. "If I believed that an international arms race was unlikely, I would be more comfortable continuing to work on this technology," he said. His advice to the public is to proactively learn about and contemplate the future direction of AI, welcoming the profound changes ahead with an open mind, and "not let others make decisions for all of us."