← Accueil actu | Salle du marché
2026-09-10 · Source：Tom's Hardware
A departing Anthropic safety researcher warns of a greater than 10 percent chance AI could kill all humans within ten years, accusing AI companies of gambling with our lives.
A safety researcher at Anthropic has put a stark number on the worst-case scenario for artificial intelligence: a greater than 10 percent chance that AI could kill all humans within the next 10 years. The warning comes not from an outside campaigner but from someone working inside a frontier AI developer — and from an employee who is now departing the company.
The figure deserves attention for both its source and its scale. A probability above 10 percent implies odds worse than one in ten that humanity does not survive the decade — an existential estimate attached to a technology being deployed commercially right now. Coming from a researcher whose role is to study AI safety, the assessment carries the weight of an insider's read on the systems being built rather than the speculation of a distant critic.
The departure adds a second dimension. In leaving, the employee accused AI companies of 'gambling with our lives' — an indictment directed at the industry's collective conduct rather than at any single firm. The framing casts the race to build ever-more-capable AI as a wager in which the stakes are human survival, and it suggests the researcher concluded that raising the alarm publicly outweighed staying inside the lab.
For the broader AI ecosystem, the intervention lands in the middle of an unresolved argument over how quickly frontier capabilities should scale and how much verification of safety claims the public should demand. Warnings from inside the labs carry particular force because they cannot be dismissed as outsider misunderstanding. Whether the more-than-10-percent estimate changes calculations at Anthropic or across the sector is unknowable from the outside, but a safety researcher walking out the door while attaching a one-in-ten odds figure to human extinction is, in itself, a measurable data point in the industry's escalating risk debate.
Mots-clés: Anthropic · AI Safety · Existential Risk · AI Doom · p(doom) · Frontier AI
萬安算交所 AIXX · AIXX · Bourse compute vérifiée · Accéder à la salle