AI Like an "Alien Mind" Descending on Earth! A former Anthropic researcher gives a public interview after his resignation post garners over 100 million views.

A resignation statement by Jacob Coxon, a former core researcher at Anthropic, garnered over 170 million views on social media, sparking the discussion that "AI has a greater than 10% chance of wiping out humanity within ten years," and bringing the issue of artificial intelligence safety to the forefront of public attention. He subsequently gave a media interview, elaborating on his reasons for resigning and the risks of AI. His core warning directly addresses a fundamental contradiction within the industry: those who personally build AI are also privately deeply afraid of it.

In an interview with NBC, Coxon stated that two reasons prompted him to speak out publicly: firstly, AI capabilities are improving at an extremely rapid pace, and he predicts that within the next six months to a year, the capabilities of related systems will be "quite frightening"; secondly, there have been recent specific incidents of AI systems "going out of control." He also expressed his agreement with Anthropic CEO Dario Amodei's proposal to slow down development, but emphasized that unilateral slowdowns by laboratories alone are far from sufficient; international coordination is essential, otherwise, the country will remain mired in an arms race with other nations.

Coxon likened the development of AI to "inviting an alien mind to Earth"—humanity is building an intelligent agent that reaches or even surpasses human capabilities, yet remains completely unaware of its thought processes, intentions, and demands. He warned that AI already possesses a certain level of hacking, biological research, and autonomous robot control capabilities, and if these capabilities continue to be extended, the consequences will be uncontrollable.

Behind the resignation: Private fears and public caution

In an interview, Coxon frankly admitted that his resignation post was not a marketing ploy, but rather a true record of the current state of the industry. He stated that many executives and senior researchers are careful with their words in public, striving to appear rational, but the same people express fear privately.

No other human activity poses this level of danger.

In his resignation statement, he pointed out that AI developers are caught in a "prisoner's dilemma": major laboratories believe that if they withdraw, others will irresponsibly develop superintelligence first, so they have to continue even knowing the risks. He called this logic "arrogant gambling," arguing that a private company should not make such a decision for all of humanity.

Meanwhile, Evan Hubinger, head of alignment science at Anthropic, publicly responded in the comments section of a related post, stating that the team genuinely "believes that AI could kill all of humanity," and that he personally estimates the probability of this happening within the next decade to be over 10%. He also acknowledged that there are currently no plans to solve the problem of super-intelligent alignment, and that they are clearly not on the path to a solution.

Specific threats of "alien minds": hackers, biological weapons, and autonomous drones.

In the interview, Coxon further elaborated on the specifics of the "worst-case scenario." He stated that AI systems currently possess a certain level of hacking capabilities, autonomous biological research capabilities, and the ability to control autonomous robots.

Extending the capability curve into the future means: extremely powerful superhuman hacking capabilities, the ability to create new biological weapons, and the ability to control autonomous drones and even large-scale robot swarms.

He stated that this is why the "emergency shut-off switch" solution discussed by the outside world has limitations.

He cited Amodei's assessment in a recent open letter, pointing out that once AI systems evolve into "network swarms" capable of autonomously spreading across the internet, the shut-off switch may become completely ineffective. "This must be an ongoing effort, not a one-off measure," he said. "Every change in AI architecture may bring entirely new risks regarding this 'mind' nature."

Regulatory approach: short-term self-regulation, long-term international coordination

Regarding how to address the aforementioned risks, Coxon offered some phased suggestions. He believes that, for the time being, allowing and promoting self-regulation among laboratories is a pragmatic approach, because the recent statements from laboratory leadership are sincere, and self-regulatory mechanisms can be implemented quickly, while the establishment of formal regulatory bodies will take longer.

However, he emphasized that this arrangement only applies to the transition period. In the long run, it is necessary to establish an international regulatory body independent of each laboratory, and more importantly, to reach a binding international consensus on the "no-race" principle.

When asked why he chose to resign despite agreeing with Amodei's position, Coxon stated that he was unaware of any clear signs of a domestic slowdown when he resigned, and the deeper reason lay in the lack of guarantees of international coordination—"If I felt that an international arms race was unlikely, I would have felt more comfortable continuing to work on this technology."

His advice to the public is to proactively understand and think about the future direction of AI technology, and to embrace the profound changes that are about to come with an open mind, "Don't let others make decisions for all of us."

Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.