AI: Moral Dilemma and Resignations at Anthropic

Employees in AI are torn between leaving their positions out of fear of extreme risks and staying to influence safety. At Anthropic, public announcements have reignited the debate, featuring risk assessments, internal support, and historical comparisons to the Manhattan Project. The American public is largely concerned.

Historical Warnings and Institutional Pressure Described

Alex Wellerstein, a professor at the Stevens Institute of Technology, believes that internal voices have little success in countering institutional momentum and considers that staying at a company equates to participating in its actions. Historian Matthew H. Hersch compares the current situation to the Manhattan Project during World War II. At that time, Ralph Austin Bard, then Undersecretary of the Navy, secretly recommended warning Japan before the use of the atomic bomb and resigned a few weeks before Hiroshima. Other scientists from the Manhattan Project had unsuccessfully attempted to secure a demonstration of the bomb before its use in combat, with some later excluded from future defense work. According to Hersch, whistleblowers can have a positive impact, but they risk being discredited or harassed into silence, which can render them unemployable in their field.

Divergent Individual Positions Within Anthropic

At Anthropic, Anna Wang explains that she works at the company because she believes she can better mitigate AI risks from the inside, while respecting those who prefer to act from the outside. A spokesperson for Anthropic confirmed her employment but declined to comment. Conversely, Jacob Coxon shared on X his decision to leave the company, stating that some AI developers genuinely believe it could lead to the death of all humanity by the end of the decade. Evan Hubinger, who leads the alignment stress testing team at Anthropic, supported Coxon's position and believes the probability of AI wiping out the entire human population in the next ten years exceeds 10%.

To Stay or Leave: Arguments from a Security Specialist

According to Jeremie Harris, CEO and co-founder of Gladstone AI, the decision to leave or stay is difficult for AI employees who are concerned about security issues. Gladstone AI, based in Washington, DC, focuses its efforts on national security dangers and catastrophic risks associated with advanced AI. Harris notes that he frequently exchanges views with employees from Anthropic and other major AI labs, and he has gathered numerous testimonies from individuals sharing the same concerns as Jacob Coxon. He points out that if departures increase, AI labs may lack collaborators who can advocate for change. He believes that those who remain are likely the least concerned about security issues, while these positions should ideally be filled by those who care the most.

Image Effects and Financial Timing Discussed

Jacob Coxon's resignation has sparked online discussions about the potential consequences for Anthropic's image. These discussions also touch on the company's upcoming IPO and the possibility of a public relations operation, drawing attention to risk management within the lab.

A Debate Fueled by Public Figures and Opinion

The debate initiated by Dario Amodei on promoting AI safety from within or outside occurs against a backdrop of alerts echoed by Geoffrey Hinton, Elon Musk, and other figures. According to a Pew Research Center study conducted in June, 52% of Americans say they are more concerned than enthusiastic about the use of AI in daily life, compared to 37% in 2021. This climate intensifies the dilemma for workers in the sector, caught between resignation and internal action to mitigate risks.