# Inside the Warning That AI Could End Humanity

> Anthropic researchers warn of a 10 percent chance AI destroys humanity within a decade, admitting they lack a plan to stop it.
- Title: Inside the Warning That AI Could End Humanity · The Verge
- Summary: Anthropic researchers warn of a 10 percent chance AI destroys humanity within a decade, admitting they lack a plan to stop it. Top safety researchers are…
- Keywords: anthropic, tech safety, existential risk, technology, culture, Inside, Warning, End, Humanity, The Verge
- Source: The Verge — https://theverge.com/ai-artificial-intelligence/991927/anthropic-ai-kill-all-humans
- Author: Robert Hart
- Published: 2026-09-09T09:56:28+00:00
- Read time: 1 min
- Topics: ai, anthropic, tech safety, existential risk, technology, culture
## Anthropic researcher quits over safety
Jacob Coxon resigned from Anthropic, accusing top AI labs of rushing recklessly toward self-improving superintelligence and risking human lives.

> racing straight to self-improving superintelligence and gambling with our lives
## A ten percent existential threat
Anthropic safety team lead Evan Hubinger agreed with the concerns, estimating a greater than 10 percent chance that AI could kill all humans within the next decade.

> We really do earnestly believe AI could kill all humans
## The dangerous loop of self-improvement
Industry insiders fear recursive self-improvement, a process where AI systems rapidly upgrade their own code in an uncontrollable feedback loop.
## No safety plan in place
Despite acknowledging the catastrophic risks, Anthropic leaders admitted they currently have no working plan to keep advanced systems aligned with human values.

> not yet have a plan
## Exodus across top AI labs
Coxon's departure highlights a broader wave of researchers exiting major AI companies as commercial pressures and upcoming IPOs overshadow safety protocols.
## Key takeaway

Top safety researchers are leaving leading AI labs, warning that the industry is racing toward dangerous superintelligence without safety safeguards.