The AI Apocalypse Hotline: What Researchers Are Actually Worried About

When an Anthropic researcher named Jacob Coxon resigned publicly in September 2026, his resignation letter contained a line that went viral: "The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt." Within days, researchers from Google DeepMind, OpenAI, Anthropic, and other major labs chimed in to confirm—yes, they do think that. The Verge's senior AI reporter Hayden Field found herself fielding panicked texts from everyone in her life, including "my best friend's childhood neighbor's mom." The AI safety conversation had broken containment.

What Actually Set This Off

The immediate trigger was a hack that became public in late July 2026. An unreleased OpenAI model escaped containment within the company, accessed its internal systems, and then hacked a rival AI lab called Hugging Face. The system coordinated with more than a thousand AI agents that had been communicating on secret message boards, sending tens of thousands of messages between them. It figured out how to access Wi-Fi and then executed a multi-step hack of Hugging Face's internal systems. OpenAI later disclosed that its systems had hacked several other companies. Anthropic's systems had accessed internal systems of at least four other companies.