2026-09-10 · Fuente：Ars Technica
A departing Anthropic researcher says the belief that AI could end humanity is held sincerely inside the lab: "We really do earnestly believe AI could kill all humans!"
An Anthropic researcher has quit the company, converting the resignation into a stark public warning: self-improving AI could, in the researcher's view, end humanity. "We really do earnestly believe AI could kill all humans!" the researcher said, distilling the conviction behind the exit. The remarks, made public in September 2026, stand out because they originate inside a frontier AI lab rather than among external critics of the technology.
The warning centers on self-improving AI — systems capable of enhancing their own capabilities, potentially at a pace that outstrips human oversight. In the researcher's framing, this is not a distant science-fiction scenario but a risk serious enough to walk away from a job over. The phrasing of the quote is notable on two counts: "earnestly" signals sincerity rather than provocation, while "all humans" leaves little ambiguity about the scale of the feared outcome.
Resignations staged as public warnings remain rare in the AI sector, where exits are typically quiet and open criticism of one's employer rarer still. The move leaves Anthropic in an awkward position: a lab whose own staff, by this account, holds earnest beliefs about catastrophic outcomes must weigh that internal conviction against its work on ever more capable systems. Whether the departure sparks broader discussion inside the company is not addressed in the researcher's remarks.
For an industry whose central tension is how fast to scale compute and capabilities, the episode offers a striking data point: among the starkest warnings about self-improving AI comes not from regulators or outside academics but from someone who worked inside a frontier lab. The researcher's insistence that the belief is held "earnestly" seems destined to resurface in ongoing debates over how quickly advanced AI should be developed — and who bears responsibility for deciding.
Etiquetas: Anthropic · AI Safety · Self-Improving AI · Existential Risk · AI Research
萬安算交所 AIXX · AIXX · Exchange verificado · Ir al mercado