More than 1 in 10 chance AI ‘could kill all humans,’ says Anthropic safety lead after colleague quits

The number itself—10 percent—is the kind of headline-grabbing estimate that usually gets dismissed as alarmism. But when it comes from Anthropic’s own safety lead, it carries a different weight. Anthropic has built its entire brand around being the “safe” AI lab, the one that puts alignment before deployment. If the people inside that institution are putting a one-in-ten chance on human extinction by 2030, the rest of the industry should probably stop rolling its eyes. The resignation that preceded this statement is arguably the more telling signal. A colleague quitting over fears that the lab and its rivals are “carelessly racing to build superhuman systems they cannot control” suggests that the public posture of safety is colliding with internal reality. This is not a lone outsider raising concerns; it’s someone who had access to the models, the training runs, and the safety protocols, and still concluded that the trajectory is untenable. That kind of departure is harder to wave off than a tweet from an AI doomer. There’s also a structural tension worth unpacking. Anthropic, like OpenAI and Google DeepMind, is caught in a commercial arms race where deployment speed is rewarded by investors, users, and national governments. Safety research that slows down release cadence becomes a liability. The resignation may reflect that dynamic: safety teams can raise red flags, but if the company’s incentives push toward shipping ever-more-capable systems, those flags get quieter. The 10 percent figure might be the safety lead’s way of saying, publicly, what internal memos could not change. What makes this story more than a PR problem is the timing. A senior safety researcher making a probabilistic extinction claim within hours of a high-profile resignation creates a narrative that the lab itself is in crisis. Whether or not the 10 percent number is defensible, the fact that it’s being said at all—and that a colleague chose to leave rather than stay and fight—points to a deeper issue: the industry’s safety infrastructure is not keeping pace with its capability curve. As the source article notes, the post was made after the resignation, but the underlying message is about control. And control is exactly what Anthropic’s competitors are also struggling to demonstrate. Source: https://www.theverge.com/ai-artificial-intelligence/991927/anthropic-ai-kill-all-humans